Using Shape instead of TensorShape.

src/TensorFlowNET.Console/MemoryBasicTest.cs

@@ -151,7 +151,7 @@ namespace Tensorflow
         public Action<int, int> Dataset
             => (epoch, iterate) =>
             {
-                TensorShape shape = (16, 32, 32, 3);
+                Shape shape = (16, 32, 32, 3);
                 var images = np.arange(shape.size).astype(np.float32).reshape(shape.dims);
                 var data_handler = new DataHandler(new DataHandlerArgs
                 {

src/TensorFlowNET.Console/MemoryFuncGraphTest.cs

@@ -22,7 +22,7 @@ namespace Tensorflow
 
         Tensor permutation(Tensor tensor)
         {
-            TensorShape shape = (8, 64, 64, 3);
+            Shape shape = (8, 64, 64, 3);
             var images = np.arange(shape.size).astype(np.float32).reshape(shape.dims);
             return tf.constant(images);
         }

src/TensorFlowNET.Console/MemoryKerasTest.cs

@@ -19,7 +19,7 @@ namespace Tensorflow
         public Action<int, int> InputLayer
             => (epoch, iterate) =>
             {
-                TensorShape shape = (32, 256, 256, 3); // 48M
+                Shape shape = (32, 256, 256, 3); // 48M
                 var images = np.arange(shape.size).astype(np.float32).reshape(shape.dims);
 
                 var inputs = keras.Input((shape.dims[1], shape.dims[2], 3));
@@ -31,7 +31,7 @@ namespace Tensorflow
         public Action<int, int> Prediction
             => (epoch, iterate) =>
             {
-                TensorShape shape = (32, 256, 256, 3); // 48M
+                Shape shape = (32, 256, 256, 3); // 48M
                 var images = np.arange(shape.size).astype(np.float32).reshape(shape.dims);
 
                 var inputs = keras.Input((shape.dims[1], shape.dims[2], 3));

src/TensorFlowNET.Core/APIs/tf.array.cs

@@ -65,7 +65,7 @@ namespace Tensorflow
         /// <param name="shape"></param>
         /// <param name="name"></param>
         /// <returns></returns>
-        public Tensor broadcast_to(Tensor input, TensorShape shape, string name = null)
+        public Tensor broadcast_to(Tensor input, Shape shape, string name = null)
             => gen_array_ops.broadcast_to(input, shape, name: name);
 
         public Tensor check_numerics(Tensor tensor, string message, string name = null)
@@ -85,7 +85,7 @@ namespace Tensorflow
                 return tf_with(ops.name_scope(name), scope =>
                 {
                     var tensor = ops.convert_to_tensor(axis, name: "concat_dim", dtype: dtypes.int32);
-                    Debug.Assert(tensor.TensorShape.ndim == 0);
+                    Debug.Assert(tensor.shape.ndim == 0);
                     return identity(values.First(), name: scope);
                 });
             }
@@ -152,7 +152,7 @@ namespace Tensorflow
         /// <param name="name"></param>
         /// <param name="conjugate"></param>
         /// <returns></returns>
-        public Tensor transpose<T1>(T1 a, TensorShape perm = null, string name = "transpose", bool conjugate = false)
+        public Tensor transpose<T1>(T1 a, Shape perm = null, string name = "transpose", bool conjugate = false)
             => array_ops.transpose(a, perm, name, conjugate);
 
         /// <summary>
@@ -246,7 +246,7 @@ namespace Tensorflow
         /// <typeparam name="T"></typeparam>
         /// <param name="input">A `Tensor`. The default value to produce when output is not fed.</param>
         /// <param name="shape">
-        /// A `tf.TensorShape` or list of `int`s. The (possibly partial) shape of
+        /// A `tf.Shape` or list of `int`s. The (possibly partial) shape of
         /// the tensor.
         /// </param>
         /// <param name="name">A name for the operation (optional).</param>

src/TensorFlowNET.Core/APIs/tf.compat.v1.cs

@@ -25,7 +25,7 @@ namespace Tensorflow
             => tf.Context.graph_mode();
 
         public IVariableV1 get_variable(string name,
-            TensorShape shape = null,
+            Shape shape = null,
             TF_DataType dtype = TF_DataType.DtInvalid,
             object initializer = null, // IInitializer or Tensor
             bool? trainable = null,

src/TensorFlowNET.Core/APIs/tf.image.cs

@@ -58,7 +58,7 @@ namespace Tensorflow
                 string name = null)
                 => image_ops_impl.resize_images(images, size, method, preserve_aspect_ratio, antialias, name);
 
-            public Tensor resize_images_v2(Tensor images, TensorShape size, string method = ResizeMethod.BILINEAR, bool preserve_aspect_ratio = false, bool antialias = false,
+            public Tensor resize_images_v2(Tensor images, Shape size, string method = ResizeMethod.BILINEAR, bool preserve_aspect_ratio = false, bool antialias = false,
                 string name = null)
                 => image_ops_impl.resize_images_v2(images, size, method, preserve_aspect_ratio, antialias, name);
 

src/TensorFlowNET.Core/APIs/tf.init.cs

@@ -96,7 +96,7 @@ namespace Tensorflow
                     seed: seed,
                     dtype: dtype);
 
-            public IInitializer zeros_initializer(TensorShape shape = null,
+            public IInitializer zeros_initializer(Shape shape = null,
                 TF_DataType dtype = TF_DataType.TF_FLOAT) => new Zeros(shape: shape,
                     dtype: dtype);
         }

src/TensorFlowNET.Core/APIs/tf.linalg.cs

@@ -27,7 +27,7 @@ namespace Tensorflow
 
             public Tensor eye(int num_rows,
                 int num_columns = -1,
-                TensorShape batch_shape = null,
+                Shape batch_shape = null,
                 TF_DataType dtype = TF_DataType.TF_DOUBLE,
                 string name = null)
                 => ops.eye(num_rows, num_columns: num_columns, batch_shape: batch_shape, dtype: dtype, name: name);

src/TensorFlowNET.Core/APIs/tf.math.cs

@@ -50,7 +50,7 @@ namespace Tensorflow
                 Tensor maxlength = null,
                 TF_DataType dtype = TF_DataType.TF_INT32,
                 string name = null,
-                TensorShape axis = null,
+                Shape axis = null,
                 bool binary_output = false)
                 => math_ops.bincount(arr, weights: weights, minlength: minlength, maxlength: maxlength,
                     dtype: dtype, name: name, axis: axis, binary_output: binary_output);
@@ -327,7 +327,7 @@ namespace Tensorflow
             => gen_math_ops.log(x, name);
 
         public Tensor equal(Tensor x, Tensor y, string name = null)
-            => gen_math_ops.equal(x, y, name);
+            => gen_math_ops.equal(x, y, name: name);
 
         /// <summary>
         /// Computes arctangent of `y/x` element-wise, respecting signs of the arguments.
@@ -453,7 +453,7 @@ namespace Tensorflow
         public static Tensor truediv(Tensor x, Tensor y, string name = null)
             => math_ops.truediv(x, y, name: name);
 
-        public Tensor range(object start, object limit = null, object delta = null, TF_DataType dtype = TF_DataType.DtInvalid, string name = "range")
+        public Tensor range(object start, object limit = null, object delta = null, TF_DataType? dtype = null, string name = "range")
             => math_ops.range(start, limit: limit, delta: delta, dtype: dtype, name: name);
 
         public Tensor real(Tensor input, string name = null)
@@ -522,7 +522,7 @@ namespace Tensorflow
                 return math_ops.reduce_sum(input, keepdims: keepdims, name: name);
         }
 
-        public Tensor reduce_sum(Tensor input, TensorShape axis, int? reduction_indices = null,
+        public Tensor reduce_sum(Tensor input, Shape axis, int? reduction_indices = null,
             bool keepdims = false, string name = null)
             => math_ops.reduce_sum(input, axis, keepdims: keepdims, name: name);
 

src/TensorFlowNET.Core/APIs/tf.queue.cs

@@ -32,7 +32,7 @@ namespace Tensorflow
         /// <returns></returns>
         public PaddingFIFOQueue PaddingFIFOQueue(int capacity,
             TF_DataType[] dtypes,
-            TensorShape[] shapes,
+            Shape[] shapes,
             string[] names = null,
             string shared_name = null,
             string name = "padding_fifo_queue")
@@ -45,7 +45,7 @@ namespace Tensorflow
 
         public PaddingFIFOQueue PaddingFIFOQueue(int capacity,
             TF_DataType dtype,
-            TensorShape shape,
+            Shape shape,
             string shared_name = null,
             string name = "padding_fifo_queue")
             => new PaddingFIFOQueue(capacity,
@@ -66,7 +66,7 @@ namespace Tensorflow
         /// <returns></returns>
         public FIFOQueue FIFOQueue(int capacity,
             TF_DataType[] dtypes,
-            TensorShape[] shapes = null,
+            Shape[] shapes = null,
             string[] names = null,
             string shared_name = null,
             string name = "fifo_queue")
@@ -79,12 +79,12 @@ namespace Tensorflow
 
         public FIFOQueue FIFOQueue(int capacity,
             TF_DataType dtype,
-            TensorShape shape = null,
+            Shape shape = null,
             string shared_name = null,
             string name = "fifo_queue")
             => new FIFOQueue(capacity,
                 new[] { dtype },
-                new[] { shape ?? new TensorShape() },
+                new[] { shape ?? Shape.Null },
                 shared_name: shared_name,
                 name: name);
 
@@ -99,26 +99,26 @@ namespace Tensorflow
         /// <returns></returns>
         public PriorityQueue PriorityQueue(int capacity,
             TF_DataType dtype,
-            TensorShape shape = null,
+            Shape shape = null,
             string shared_name = null,
             string name = "priority_queue")
             => new PriorityQueue(capacity,
                 new[] { dtype },
-                new[] { shape ?? new TensorShape() },
+                new[] { shape ?? Shape.Null },
                 shared_name: shared_name,
                 name: name);
 
         public RandomShuffleQueue RandomShuffleQueue(int capacity,
             int min_after_dequeue,
             TF_DataType dtype,
-            TensorShape shape = null,
+            Shape shape = null,
             int? seed = null,
             string shared_name = null,
             string name = "random_shuffle_queue")
             => new RandomShuffleQueue(capacity,
                 min_after_dequeue: min_after_dequeue,
                 new[] { dtype },
-                new[] { shape ?? new TensorShape() },
+                new[] { shape ?? Shape.Null },
                 seed: seed,
                 shared_name: shared_name,
                 name: name);

src/TensorFlowNET.Core/APIs/tf.random.cs

@@ -32,7 +32,7 @@ namespace Tensorflow
             /// <param name="seed"></param>
             /// <param name="name"></param>
             /// <returns></returns>
-            public Tensor normal(TensorShape shape,
+            public Tensor normal(Shape shape,
                 float mean = 0.0f,
                 float stddev = 1.0f,
                 TF_DataType dtype = TF_DataType.TF_FLOAT,
@@ -49,7 +49,7 @@ namespace Tensorflow
             /// <param name="seed"></param>
             /// <param name="name"></param>
             /// <returns></returns>
-            public Tensor truncated_normal(TensorShape shape,
+            public Tensor truncated_normal(Shape shape,
                 float mean = 0.0f,
                 float stddev = 1.0f,
                 TF_DataType dtype = TF_DataType.TF_FLOAT,
@@ -63,7 +63,7 @@ namespace Tensorflow
                 string name = null,
                 TF_DataType output_dtype = TF_DataType.DtInvalid) => random_ops.multinomial(logits, num_samples, seed: seed, name: name, output_dtype: output_dtype);
 
-            public Tensor uniform(TensorShape shape,
+            public Tensor uniform(Shape shape,
                 float minval = 0,
                 float maxval = 1,
                 TF_DataType dtype = TF_DataType.TF_FLOAT,
@@ -77,7 +77,7 @@ namespace Tensorflow
             }
         }
 
-        public Tensor random_uniform(TensorShape shape,
+        public Tensor random_uniform(Shape shape,
             float minval = 0,
             float maxval = 1,
             TF_DataType dtype = TF_DataType.TF_FLOAT,
@@ -85,7 +85,7 @@ namespace Tensorflow
             string name = null)
             => random.uniform(shape, minval: minval, maxval: maxval, dtype: dtype, seed: seed, name: name);
 
-        public Tensor truncated_normal(TensorShape shape,
+        public Tensor truncated_normal(Shape shape,
             float mean = 0.0f,
             float stddev = 1.0f,
             TF_DataType dtype = TF_DataType.TF_FLOAT,

src/TensorFlowNET.Core/APIs/tf.reshape.cs

@@ -19,7 +19,7 @@ namespace Tensorflow
     public partial class tensorflow
     {
         public Tensor reshape(Tensor tensor,
-                TensorShape shape,
+                Shape shape,
                 string name = null) 
             => gen_array_ops.reshape(tensor, shape, name);
 

src/TensorFlowNET.Core/APIs/tf.sparse.cs

@@ -47,7 +47,7 @@ namespace Tensorflow
         /// <param name="name"></param>
         /// <returns>Dense `Tensor` of shape `output_shape`.  Has the same type as `sparse_values`.</returns>
         public Tensor sparse_to_dense<T>(Tensor sparse_indices,
-            TensorShape output_shape,
+            Shape output_shape,
             T sparse_values,
             T default_value = default,
             bool validate_indices = true,

src/TensorFlowNET.Core/APIs/tf.tile.cs

@@ -25,7 +25,7 @@ namespace Tensorflow
         public Tensor tile(Tensor input, object[] multiples, string name = null)
             => gen_array_ops.tile(input, multiples, name);
 
-        public Tensor tile(Tensor input, TensorShape multiples, string name = null) 
+        public Tensor tile(Tensor input, Shape multiples, string name = null) 
         {
             var multiples_tensor = constant_op.constant(multiples);
             return gen_array_ops.tile(input, multiples_tensor, name);

src/TensorFlowNET.Core/Binding.Util.cs

@@ -168,7 +168,7 @@ namespace Tensorflow
                     return ndArray.ndim == 0 ? 1 : (int)ndArray.dims[0];
                 case IEnumerable enumerable:
                     return enumerable.OfType<object>().Count();
-                case TensorShape arr:
+                case Shape arr:
                     return arr.ndim;
             }
             throw new NotImplementedException("len() not implemented for type: " + a.GetType());
@@ -532,7 +532,6 @@ namespace Tensorflow
             var type = data.GetType();
             switch (data)
             {
-                case TensorShape:
                 case Shape:
                     return TF_DataType.TF_INT64;
                 case Axis:

src/TensorFlowNET.Core/Binding.cs

@@ -8,12 +8,12 @@ namespace Tensorflow
 
         /// <summary>
         ///     Alias to null, similar to python's None.
-        ///     For TensorShape, please use Unknown
+        ///     For Shape, please use Unknown
         /// </summary>
         public static readonly object None = null;
 
         /// <summary>
-        /// Used for TensorShape None
+        /// Used for Shape None
         /// </summary>
         /// 
         public static readonly int Unknown = -1;

src/TensorFlowNET.Core/Data/DatasetManager.cs

@@ -6,7 +6,7 @@ namespace Tensorflow
 {
     public class DatasetManager
     {
-        public IDatasetV2 from_generator<T>(IEnumerable<T> generator, TF_DataType[] output_types, TensorShape[] output_shapes)
+        public IDatasetV2 from_generator<T>(IEnumerable<T> generator, TF_DataType[] output_types, Shape[] output_shapes)
             => new GeneratorDataset();
 
         /// <summary>

src/TensorFlowNET.Core/Data/DatasetV2.cs

@@ -19,7 +19,7 @@ namespace Tensorflow
 
         public TensorSpec[] structure { get; set; }
 
-        public TensorShape[] output_shapes => structure.Select(x => x.shape).ToArray();
+        public Shape[] output_shapes => structure.Select(x => x.shape).ToArray();
 
         public TF_DataType[] output_types => structure.Select(x => x.dtype).ToArray();
 

src/TensorFlowNET.Core/Data/IDatasetV2.cs

@@ -10,7 +10,7 @@ namespace Tensorflow
 
         Tensor variant_tensor { get; set; }
 
-        TensorShape[] output_shapes { get; }
+        Shape[] output_shapes { get; }
 
         TF_DataType[] output_types { get; }
 

src/TensorFlowNET.Core/Data/OwnedIterator.cs

@@ -37,7 +37,7 @@ namespace Tensorflow
             {
                 var results = ops.iterator_get_next(_iterator_resource, _dataset.output_types, _dataset.output_shapes);
                 foreach(var (i, tensor) in enumerate(results))
-                    tensor.set_shape(_element_spec[i].shape);
+                    tensor.shape = _element_spec[i].shape;
                 return results;
             }
             catch (OutOfRangeError ex)

src/TensorFlowNET.Core/Eager/EagerRunner.TFE_FastPathExecute.cs

@@ -298,7 +298,7 @@ namespace Tensorflow.Eager
                 c_api.TFE_OpSetAttrStringList(op, key, values3, values3.Select(x => Convert.ToUInt64(x.Length)).ToArray(), values3.Length);
                 attr_list_sizes[key] = values3.Length;
             }
-            else if (type == TF_AttrType.TF_ATTR_SHAPE && values is TensorShape[] values1)
+            else if (type == TF_AttrType.TF_ATTR_SHAPE && values is Shape[] values1)
             {
                 // Make one pass through the input counting the total number of
                 // dims across all the input lists.

src/TensorFlowNET.Core/Eager/EagerTensor.ToString.cs

@@ -3,6 +3,6 @@
     public partial class EagerTensor
     {
         public override string ToString()
-            => $"tf.Tensor: shape={TensorShape}, dtype={dtype.as_numpy_name()}, numpy={tensor_util.to_numpy_string(this)}";
+            => $"tf.Tensor: shape={shape}, dtype={dtype.as_numpy_name()}, numpy={tensor_util.to_numpy_string(this)}";
     }
 }

src/TensorFlowNET.Core/Eager/EagerTensor.cs

@@ -17,12 +17,6 @@ namespace Tensorflow.Eager
 
         public override int rank => c_api.TFE_TensorHandleNumDims(EagerTensorHandle, tf.Status.Handle);
 
-        public override void set_shape(TensorShape shape)
-        {
-            if (!shape.is_compatible_with(this.shape))
-                throw new ValueError($"Tensor's shape is not compatible.");
-        }
-
         public static int GetRank(IntPtr handle)
         {
             var tfe_tensor_handle = c_api.TFE_EagerTensorHandle(handle);

src/TensorFlowNET.Core/Framework/Models/DenseSpec.cs

@@ -5,8 +5,8 @@
     /// </summary>
     public class DenseSpec : TypeSpec
     {
-        protected TensorShape _shape;
-        public TensorShape shape => _shape;
+        protected Shape _shape;
+        public Shape shape => _shape;
 
         protected TF_DataType _dtype;
         public TF_DataType dtype => _dtype;
@@ -14,7 +14,7 @@
         protected string _name;
         public string name => _name;
 
-        public DenseSpec(TensorShape shape, TF_DataType dtype = TF_DataType.TF_FLOAT, string name = null)
+        public DenseSpec(Shape shape, TF_DataType dtype = TF_DataType.TF_FLOAT, string name = null)
         {
             _shape = shape;
             _dtype = dtype;

src/TensorFlowNET.Core/Framework/Models/TensorSpec.cs

@@ -4,7 +4,7 @@ namespace Tensorflow.Framework.Models
 {
     public class TensorSpec : DenseSpec
     {
-        public TensorSpec(TensorShape shape, TF_DataType dtype = TF_DataType.TF_FLOAT, string name = null) :
+        public TensorSpec(Shape shape, TF_DataType dtype = TF_DataType.TF_FLOAT, string name = null) :
             base(shape, dtype, name)
         {
 

src/TensorFlowNET.Core/Framework/common_shapes.py.cs

@@ -34,8 +34,8 @@ namespace Tensorflow.Framework
         /// <summary>
         /// Helper functions for is_broadcast_compatible and broadcast_shape.
         /// </summary>
-        /// <param name="shape_x"> A `TensorShape`</param>
-        /// <param name="shape_y"> A `TensorShape`</param>
+        /// <param name="shape_x"> A `Shape`</param>
+        /// <param name="shape_y"> A `Shape`</param>
         /// <return> Returns None if the shapes are not broadcast compatible,
         /// a list of the broadcast dimensions otherwise.
         /// </return>
@@ -51,7 +51,7 @@ namespace Tensorflow.Framework
 
         public static bool has_fully_defined_shape(Tensor tensor)
         {
-            return tensor.TensorShape.is_fully_defined();
+            return tensor.shape.IsFullyDefined;
         }
     }
 }

src/TensorFlowNET.Core/Framework/tensor_shape.cs

@@ -28,7 +28,7 @@ namespace Tensorflow.Framework
         {
             bool _shape_is_compatible_0dim(Shape _this, Shape _other)
             {
-                var __other = tensor_shape.as_shape(_other);
+                var __other = _other;
                 if (_this.dims == null || __other.dims == null)
                     return true;
 
@@ -54,9 +54,9 @@ namespace Tensorflow.Framework
                    !self.IsSparseTensor;
         }
 
-        public static Dimension dimension_at_index(TensorShape shape, int index)
+        public static Dimension dimension_at_index(Shape shape, int index)
         {
-            return shape.rank < 0 ?
+            return shape.ndim < 0 ?
                 new Dimension(-1) :
                 new Dimension(shape.dims[index]);
         }
@@ -64,19 +64,16 @@ namespace Tensorflow.Framework
         public static int dimension_value(Dimension dimension)
             => (int)dimension.value;
 
-        public static TensorShape as_shape(this Shape shape)
-             => new TensorShape(shape.dims);
-
-        public static TensorShape most_specific_compatible_shape(this TensorShape self, TensorShape other)
+        public static Shape most_specific_compatible_shape(this Shape self, Shape other)
         {
-            var dims = range(self.rank).Select(x => -1L).ToArray();
+            var dims = range(self.ndim).Select(x => -1L).ToArray();
             foreach(var (i, (d1, d2)) in enumerate(zip(self.dims, other.dims)))
             {
                 if (d1 == d2)
                     dims[i] = d1;
             }
 
-            return new TensorShape(dims);
+            return new Shape(dims);
         }
     }
 }

src/TensorFlowNET.Core/Functions/ConcreteFunction.cs

@@ -72,7 +72,7 @@ namespace Tensorflow.Functions
         }
 
         /*public ConcreteFunction(Func<Tensors, Tensors> func,
-            TF_DataType[] dtypes, TensorShape[] shapes)
+            TF_DataType[] dtypes, Shape[] shapes)
         {
             string func_name = $"{func.Method.Name}_{ops.uid_function()}";
 

src/TensorFlowNET.Core/Gradients/TapeTensor.cs

@@ -6,9 +6,9 @@ namespace Tensorflow.Gradients
     {
         long id;
         TF_DataType dtype;
-        TensorShape shape;
+        Shape shape;
 
-        public TapeTensor(long id, TF_DataType dtype, TensorShape shape)
+        public TapeTensor(long id, TF_DataType dtype, Shape shape)
         {
             this.id = id;
             this.dtype = dtype;

src/TensorFlowNET.Core/Gradients/gradients_util.cs

@@ -234,7 +234,7 @@ namespace Tensorflow
                                         in_grad.Tag == null && // maybe a IndexedSlice
                                         t_in.dtype != TF_DataType.TF_RESOURCE)
                                     {
-                                        in_grad.set_shape(t_in.TensorShape);
+                                        in_grad.shape = t_in.shape;
                                     }
 
                                     _SetGrad(grads, t_in, in_grad);

src/TensorFlowNET.Core/Gradients/image_grad.cs

@@ -27,9 +27,9 @@ namespace Tensorflow.Gradients
         {
             var grad = grads[0];
             var image = op.inputs[0];
-            var shape = new TensorShape(image.shape.dims.Skip(1).Take(2).ToArray());
+            var shape = new Shape(image.shape.dims.Skip(1).Take(2).ToArray());
             Tensor image_shape = null;
-            if (shape.is_fully_defined())
+            if (shape.IsFullyDefined)
                 image_shape = constant_op.constant(image.shape.dims.Skip(1).Take(2).ToArray());
             else
                 image_shape = array_ops.shape(image)["1:3"];

src/TensorFlowNET.Core/Gradients/math_grad.cs

@@ -810,8 +810,8 @@ namespace Tensorflow.Gradients
         private static (Tensor, Tensor, bool)[] SmartBroadcastGradientArgs(Tensor x, Tensor y, Tensor grad)
         {
             Tensor sx, sy;
-            if (x.TensorShape.is_fully_defined() &&
-                y.TensorShape.is_fully_defined())
+            if (x.shape.IsFullyDefined &&
+                y.shape.IsFullyDefined)
             {
                 sx = array_ops.shape(x);
                 sy = array_ops.shape(y);
@@ -825,8 +825,8 @@ namespace Tensorflow.Gradients
             var (rx, ry) = gen_array_ops.broadcast_gradient_args(sx, sy);
             return new[]
             {
-                (sx, rx, !x.TensorShape.Equals(grad.TensorShape)),
-                (sy, ry, !y.TensorShape.Equals(grad.TensorShape))
+                (sx, rx, !x.shape.Equals(grad.shape)),
+                (sy, ry, !y.shape.Equals(grad.shape))
             };
         }
     }

src/TensorFlowNET.Core/Gradients/nn_grad.cs

@@ -17,6 +17,7 @@
 using System;
 using System.Linq;
 using Tensorflow.Operations;
+using static Tensorflow.Binding;
 
 namespace Tensorflow.Gradients
 {
@@ -323,7 +324,7 @@ namespace Tensorflow.Gradients
 
             // Compute linear indices(flattened to 1D).
             var cast1 = math_ops.cast(outerdim, TF_DataType.TF_INT64);
-            var range2 = math_ops.range(0L, cast1 * in_lastdim, in_lastdim);
+            var range2 = math_ops.range(tf.constant(0L), cast1 * in_lastdim, in_lastdim);
             var dim2 = array_ops.expand_dims(range2, -1);
             var cast2 = math_ops.cast(dim2, TF_DataType.TF_INT32);
             var ind = array_ops.reshape(ind_2d + cast2, new int[] { -1 });

src/TensorFlowNET.Core/Gradients/ops.gradient_function_mapping.cs

@@ -56,7 +56,7 @@ namespace Tensorflow
                                    null,
                                    args: new object[] { oper, out_grads }) as Tensor[];
                                 foreach (var result in results.Where(x => x != null))
-                                    tf.Logger.Debug($"Gradient: {result.name} {result.TensorShape}");
+                                    tf.Logger.Debug($"Gradient: {result.name} {result.shape}");
                                 return results;
                             }
                         );

src/TensorFlowNET.Core/Graphs/AutoGraphAttribute.cs

@@ -44,7 +44,7 @@ namespace Tensorflow.Graphs
             if (args.Arguments[0] is Tensors inputs)
             {
                 originalInputs = inputs;
-                var new_inputs = inputs.Select(x => tf.placeholder(x.dtype, shape: x.TensorShape, name: "inputs")).ToArray();
+                var new_inputs = inputs.Select(x => tf.placeholder(x.dtype, shape: x.shape, name: "inputs")).ToArray();
                 args.Arguments[0] = new Tensors(new_inputs);
             }
             else
@@ -56,7 +56,7 @@ namespace Tensorflow.Graphs
                     if (args.Arguments[i] is EagerTensor tensor)
                     {
                         originalInputs.Add(tensor);
-                        args.Arguments[i] = tf.placeholder(tensor.dtype, shape: tensor.TensorShape, name: "inputs");
+                        args.Arguments[i] = tf.placeholder(tensor.dtype, shape: tensor.shape, name: "inputs");
                     }
                 }
             }

src/TensorFlowNET.Core/Graphs/FuncGraph.cs

@@ -103,7 +103,7 @@ namespace Tensorflow.Graphs
         }
 
         const int _EAGER_CONST_THRESHOLD = 128;
-        public Tensor capture(Tensor tensor, string name = null, TensorShape shape = null)
+        public Tensor capture(Tensor tensor, string name = null, Shape shape = null)
         {
             if(tensor is EagerTensor)
             {
@@ -167,7 +167,7 @@ namespace Tensorflow.Graphs
             return graph_const;
         }
 
-        Tensor _capture_helper(Tensor tensor, string name, TensorShape shape = null)
+        Tensor _capture_helper(Tensor tensor, string name, Shape shape = null)
         {
             Tensor placeholder = null;
             if (!_captures.ContainsKey(tensor.Id))
@@ -206,7 +206,7 @@ namespace Tensorflow.Graphs
         Tensor _create_substitute_placeholder(Tensor value, 
             string name = null, 
             TF_DataType dtype = TF_DataType.DtInvalid, 
-            TensorShape shape = null)
+            Shape shape = null)
         {
             if (shape is null)
                 shape = value.shape;

src/TensorFlowNET.Core/Graphs/Graph.cs

@@ -515,20 +515,20 @@ namespace Tensorflow
             return (Tensor)this.as_graph_element(name, allow_tensor: true, allow_operation: false);
         }
 
-        public TensorShape GetTensorShape(TF_Output output)
+        public Shape GetTensorShape(TF_Output output)
         {
             var status = tf.Status;
             var ndim = c_api.TF_GraphGetTensorNumDims(_handle, output, status.Handle);
             status.Check();
 
             if (ndim == -1)
-                return new TensorShape();
+                return Shape.Null;
 
             var dims = new long[ndim];
             c_api.TF_GraphGetTensorShape(_handle, output, dims, dims.Length, status.Handle);
             status.Check();
 
-            return new TensorShape(dims.Select(x => (int)x).ToArray());
+            return new Shape(dims.Select(x => (int)x).ToArray());
         }
 
         public virtual void Exit()

src/TensorFlowNET.Core/Keras/ArgsDefinition/Convolution/ConvolutionalArgs.cs

@@ -8,16 +8,16 @@ namespace Tensorflow.Keras.ArgsDefinition
         public int Rank { get; set; } = 2;
         public int Filters { get; set; }
         public int NumSpatialDims { get; set; } = Unknown;
-        public TensorShape KernelSize { get; set; } = 5;
+        public Shape KernelSize { get; set; } = 5;
 
         /// <summary>
         /// specifying the stride length of the convolution.
         /// </summary>
-        public TensorShape Strides { get; set; } = (1, 1);
+        public Shape Strides { get; set; } = (1, 1);
 
         public string Padding { get; set; } = "valid";
         public string DataFormat { get; set; }
-        public TensorShape DilationRate { get; set; } = (1, 1);
+        public Shape DilationRate { get; set; } = (1, 1);
         public int Groups { get; set; } = 1;
         public Activation Activation { get; set; }
         public bool UseBias { get; set; }

src/TensorFlowNET.Core/Keras/ArgsDefinition/LayerArgs.cs

@@ -25,12 +25,12 @@
         /// <summary>
         /// Only applicable to input layers.
         /// </summary>
-        public TensorShape InputShape { get; set; }
+        public Shape InputShape { get; set; }
 
         /// <summary>
         /// Only applicable to input layers.
         /// </summary>
-        public TensorShape BatchInputShape { get; set; }
+        public Shape BatchInputShape { get; set; }
 
         public int BatchSize { get; set; } = -1;
 

src/TensorFlowNET.Core/Keras/ArgsDefinition/Normalization/BatchNormalizationArgs.cs

@@ -4,7 +4,7 @@ namespace Tensorflow.Keras.ArgsDefinition
 {
     public class BatchNormalizationArgs : LayerArgs
     {
-        public TensorShape Axis { get; set; } = -1;
+        public Shape Axis { get; set; } = -1;
         public float Momentum { get; set; } = 0.99f;
         public float Epsilon { get; set; } = 1e-3f;
         public bool Center { get; set; } = true;

src/TensorFlowNET.Core/Keras/ArgsDefinition/Pooling/Pooling2DArgs.cs

@@ -10,12 +10,12 @@
         /// <summary>
         /// specifying the size of the pooling window.
         /// </summary>
-        public TensorShape PoolSize { get; set; }
+        public Shape PoolSize { get; set; }
 
         /// <summary>
         /// specifying the strides of the pooling operation.
         /// </summary>
-        public TensorShape Strides { get; set; }
+        public Shape Strides { get; set; }
 
         /// <summary>
         /// The padding method, either 'valid' or 'same'.

src/TensorFlowNET.Core/Keras/ArgsDefinition/Regularization/DropoutArgs.cs

@@ -11,7 +11,7 @@
         /// 1D integer tensor representing the shape of the
         /// binary dropout mask that will be multiplied with the input.
         /// </summary>
-        public TensorShape NoiseShape { get; set; }
+        public Shape NoiseShape { get; set; }
 
         /// <summary>
         /// random seed.

src/TensorFlowNET.Core/Keras/ArgsDefinition/Reshaping/ReshapeArgs.cs

@@ -2,7 +2,7 @@
 {
     public class ReshapeArgs : LayerArgs
     {
-        public TensorShape TargetShape { get; set; }
+        public Shape TargetShape { get; set; }
         public object[] TargetShapeObjects { get; set; }
     }
 }

src/TensorFlowNET.Core/Keras/ArgsDefinition/Reshaping/UpSampling2DArgs.cs

@@ -2,7 +2,7 @@
 {
     public class UpSampling2DArgs : LayerArgs
     {
-        public TensorShape Size { get; set; }
+        public Shape Size { get; set; }
         public string DataFormat { get; set; }
         /// <summary>
         /// 'nearest', 'bilinear'

src/TensorFlowNET.Core/Keras/Engine/InputSpec.cs

@@ -27,14 +27,14 @@ namespace Tensorflow.Keras.Engine
         public int? ndim;
         public int? min_ndim;
         Dictionary<int, int> axes;
-        TensorShape shape;
+        Shape shape;
         public int[] AllAxisDim;
 
         public InputSpec(TF_DataType dtype = TF_DataType.DtInvalid,
             int? ndim = null,
             int? min_ndim = null,
             Dictionary<int, int> axes = null,
-            TensorShape shape = null)
+            Shape shape = null)
         {
             this.ndim = ndim;
             if (axes == null)

src/TensorFlowNET.Core/Keras/Layers/ILayer.cs

@@ -16,8 +16,8 @@ namespace Tensorflow.Keras
         List<IVariableV1> trainable_variables { get; }
         List<IVariableV1> trainable_weights { get; }
         List<IVariableV1> non_trainable_weights { get; }
-        TensorShape output_shape { get; }
-        TensorShape BatchInputShape { get; }
+        Shape output_shape { get; }
+        Shape BatchInputShape { get; }
         TF_DataType DType { get; }
         int count_params();
         LayerArgs get_config();

src/TensorFlowNET.Core/NumPy/Implementation/NumPyImpl.Creation.cs

@@ -28,7 +28,7 @@ namespace Tensorflow.NumPy
                     diag_len = N + k;
             }
 
-            var diagonal_ = array_ops.ones(new TensorShape(diag_len), dtype: dtype);
+            var diagonal_ = array_ops.ones(new Shape(diag_len), dtype: dtype);
             var tensor = array_ops.matrix_diag(diagonal: diagonal_, num_rows: N, num_cols: M.Value, k: k);
             return new NDArray(tensor);
         }
@@ -38,13 +38,13 @@ namespace Tensorflow.NumPy
         {
             var start_tensor = array_ops.constant(start, dtype: dtype);
             var stop_tensor = array_ops.constant(stop, dtype: dtype);
-            var num_tensor = array_ops.constant(num);
+
             // var step_tensor = array_ops.constant(np.nan);
             Tensor result = null;
 
             if (endpoint)
             {
-                result = math_ops.linspace(start_tensor, stop_tensor, num_tensor, axis: axis);
+                result = math_ops.linspace(start_tensor, stop_tensor, num, axis: axis);
             }
             else
             {
@@ -53,10 +53,10 @@ namespace Tensorflow.NumPy
                     var step = (stop_tensor - start_tensor) / num;
                     var new_stop = math_ops.cast(stop_tensor, step.dtype) - step;
                     start_tensor = math_ops.cast(start_tensor, new_stop.dtype);
-                    result = math_ops.linspace(start_tensor, new_stop, num_tensor, axis: axis);
+                    result = math_ops.linspace(start_tensor, new_stop, num, axis: axis);
                 }
                 else
-                    result = math_ops.linspace(start_tensor, stop_tensor, num_tensor, axis: axis);
+                    result = math_ops.linspace(start_tensor, stop_tensor, num, axis: axis);
             }
 
             return new NDArray(result);

src/TensorFlowNET.Core/Numpy/Shape.cs

@@ -7,16 +7,27 @@ namespace Tensorflow
 {
     public class Shape
     {
-        public int ndim => _dims.Length;
+        public int ndim => _dims == null ? -1 : _dims.Length;
         long[] _dims;
         public long[] dims => _dims;
 
-        public Shape() 
+        private Shape() 
         {
         }
 
+        public Shape(TensorShapeProto proto)
+        {
+            _dims = proto.Dim.Select(x => x.Size).ToArray();
+        }
+
+        public void Deconstruct(out long h, out long w)
+        {
+            h = dims[0];
+            w = dims[1];
+        }
+
         public Shape(params int[] dims)
-            => _dims = dims.Select(x => Convert.ToInt64(x)).ToArray();
+            => _dims = dims?.Select(x => Convert.ToInt64(x))?.ToArray();
 
         public Shape(params long[] dims)
             => _dims = dims;
@@ -25,10 +36,10 @@ namespace Tensorflow
             => new Shape(dims);
 
         public static implicit operator Shape(long[] dims)
-            => new Shape(dims);
+            => dims == null ? null : new Shape(dims);
 
         public static implicit operator Shape(int[] dims)
-            => new Shape(dims);
+            => dims == null ? null : new Shape(dims);
 
         public static implicit operator Shape((int, int) dims)
             => new Shape(dims.Item1, dims.Item2);
@@ -57,16 +68,39 @@ namespace Tensorflow
         public bool IsEmpty => size == 0;
 
         public bool IsScalar => ndim == 0;
+        public bool IsNull => _dims == null;
+
+        public bool IsFullyDefined => ndim > -1 && dims.Count(x => x < 1) == 0;
+
+        public static Shape Scalar => new Shape(new long[0]);
+        public static Shape Null => new Shape();
+
+        public long this[int n] 
+        {
+            get => dims[n];
+            set => dims[n] = value;
+        }
+
+        public Shape this[Slice slice]
+        {
+            get
+            {
+                if (!slice.Stop.HasValue)
+                    slice.Stop = dims.Length - slice.Start + 1;
 
-        public static Shape Scalar
-            => new Shape(new long[0]);
+                if (slice.Start.HasValue == false || slice.Length.HasValue == false)
+                    throw new ArgumentException("Slice must has Start and Length.");
 
-        public long this[int n] => dims[n];
+                return new Shape(dims.Skip(slice.Start.Value)
+                    .Take(slice.Length.Value)
+                    .ToArray());
+            }
+        }
 
         /// <summary>
         ///     Returns the size this shape represents.
         /// </summary>
-        public ulong size
+        public long size
         {
             get
             {
@@ -85,44 +119,133 @@ namespace Tensorflow
                     computed *= val;
                 }
 
-                return (ulong)computed;
+                return computed;
             }
         }
 
-        public bool is_fully_defined()
-        {
-            return ndim > -1 && dims != null && dims.Count(x => x < 1) == 0;
-        }
-
-        public bool is_compatible_with(TensorShape shape2)
+        public bool is_compatible_with(Shape shape2)
         {
             if (dims != null && shape2.dims != null)
             {
                 if (dims.Contains(-1) || shape2.dims.Contains(-1))
                     return true;
 
-                if (size != (ulong)shape2.size)
+                if (size != shape2.size)
                     return false;
             }
 
             return true;
         }
 
+        public Shape with_rank_at_least(int rank)
+        {
+            if (ndim < rank)
+                throw new ValueError($"Shape {this} must have rank at least {rank}");
+            else
+                return this;
+        }
+
+        public Shape with_rank(int rank)
+        {
+            return merge_with(unknown_shape(rank: rank));
+        }
+
+        /// <summary>
+        /// Returns an unknown Shape, optionally with a known rank.
+        /// </summary>
+        /// <param name="rank"></param>
+        /// <returns></returns>
+        public Shape unknown_shape(int rank = -1)
+        {
+            if (rank == -1)
+                return Shape.Null;
+            else
+                return new Shape(Enumerable.Repeat(-1L, rank).ToArray());
+        }
+
+        public Shape concatenate(long[] other)
+        {
+            return concatenate(new Shape(other));
+        }
+
+        /// <summary>
+        ///     Returns the concatenation of the dimension in `self` and `other`.
+        /// </summary>
+        /// <param name="other"></param>
+        /// <returns></returns>
+        public Shape concatenate(Shape other)
+        {
+            var otherShape = other;
+
+            if (ndim < 0 || otherShape.ndim < 0)
+                return Shape.Null;
+            else
+            {
+                var concatenate_dims = new long[ndim + otherShape.ndim];
+                for (int i = 0; i < ndim; i++)
+                    concatenate_dims[i] = dims[i];
+
+                for (int i = 0; i < otherShape.ndim; i++)
+                    concatenate_dims[ndim + i] = otherShape.dims[i];
+
+                return new Shape(concatenate_dims);
+            }
+        }
+
+        /// <summary>
+        /// Returns a `Shape` combining the information in `self` and `other`.
+        /// </summary>
+        /// <param name="other"></param>
+        /// <returns></returns>
+        public Shape merge_with(Shape other)
+        {
+            if (dims == null)
+                return other;
+
+            var new_dims = new List<long>();
+
+            foreach (var i in Enumerable.Range(0, ndim))
+            {
+                var dim = new Dimension(dims[i]);
+                var merged = dim.merge_with(new Dimension(other.dims[i]));
+                new_dims.Add(merged.value);
+            }
+
+            return new Shape(new_dims.ToArray());
+        }
+
+        public void assert_has_rank(int rank)
+        {
+            if (rank != ndim)
+                throw new ValueError(String.Format("Shape {0} must have rank {1}", ndim, rank));
+        }
+
         public override bool Equals(object obj)
         {
-            if(obj is Shape shape)
+            switch (obj)
             {
-                if (shape.ndim != ndim)
+                case Shape shape1:
+                    if (ndim == -1 && shape1.ndim == -1)
+                        return false;
+                    else if (ndim != shape1.ndim)
+                        return false;
+                    return Enumerable.SequenceEqual(shape1.dims, dims);
+                case long[] shape2:
+                    if (ndim != shape2.Length)
+                        return false;
+                    return Enumerable.SequenceEqual(dims, shape2);
+                default:
                     return false;
-                if (Enumerable.SequenceEqual(dims, shape.dims))
-                    return true;
             }
-            return base.Equals(obj);
         }
 
         public override string ToString()
-        {
-            return "(" + string.Join(", ", _dims) + ")";
-        }
+            => ndim switch
+            {
+                -1 => "<unknown>",
+                0 => "()",
+                1 => $"({dims[0]},)",
+                _ => $"{string.Join(", ", _dims).Replace("-1", "None")}"
+            };
     }
 }

src/TensorFlowNET.Core/Operations/ControlFlows/ControlFlowState.cs

@@ -261,7 +261,7 @@ namespace Tensorflow.Operations.ControlFlows
         public Tensor ZerosLikeForExit(Tensor val)
         {
             Tensor result = null;
-            var val_shape = val.TensorShape;
+            var val_shape = val.shape;
             var forward_ctxt = val.op._get_control_flow_context();
             var outer_forward_ctxt = forward_ctxt.outer_context;
             if (outer_forward_ctxt != null)
@@ -278,7 +278,7 @@ namespace Tensorflow.Operations.ControlFlows
             {
                 // If the shape is known statically, just create a zero tensor
                 // with the right shape.
-                if (val_shape.is_fully_defined())
+                if (val_shape.IsFullyDefined)
                     result = array_ops.zeros(val_shape.dims, val.dtype);
                 else
                     result = array_ops.zeros_like(val, optimize: false);
@@ -299,8 +299,8 @@ namespace Tensorflow.Operations.ControlFlows
                         // depend on its value at iteration i. So use zeros as the
                         // gradients for all iterations > 0.
                         var dtype = b_merge.op.inputs[0].dtype;
-                        var shape = b_merge.op.inputs[0].TensorShape;
-                        if (shape.is_fully_defined())
+                        var shape = b_merge.op.inputs[0].shape;
+                        if (shape.IsFullyDefined)
                         {
                             grad_state.grad_context.Enter();
                             // Create a zeros and use it for iterations > 0.

src/TensorFlowNET.Core/Operations/ControlFlows/GradLoopState.cs

@@ -258,7 +258,7 @@ namespace Tensorflow.Operations.ControlFlows
                     throw new NotImplementedException("AddBackpropAccumulatedValue");
                 }
                 pop = gen_data_flow_ops.stack_pop_v2(history_value, value.dtype.as_base_dtype());
-                pop.set_shape(value.TensorShape);
+                pop.shape = value.shape;
                 grad_context.Exit();
             });
             var parallel_iterations = grad_context.parallel_iterations;

src/TensorFlowNET.Core/Operations/ControlFlows/WhileContext.cs

@@ -117,7 +117,7 @@ namespace Tensorflow.Operations
         internal LoopVar<TItem> BuildLoop<TItem>(Func<LoopVar<TItem>, Tensor> pred,
             Func<LoopVar<TItem>, LoopVar<TItem>> body,
             LoopVar<TItem> loop_vars,
-            TensorShape[] shape_invariants,
+            Shape[] shape_invariants,
             bool return_same_structure) where TItem : IFromMergeVars<TItem>, new()
         {
             // Keep original_loop_vars to identify which are TensorArrays
@@ -159,9 +159,9 @@ namespace Tensorflow.Operations
             throw new NotImplementedException("_convert_tensorarray_to_flow");
         }
 
-        private TensorShape _get_shape_invariant(Tensor var, int[] shape = null)
+        private Shape _get_shape_invariant(Tensor var, int[] shape = null)
         {
-            return var.TensorShape;
+            return var.shape;
         }
 
         /// <summary>
@@ -178,7 +178,7 @@ namespace Tensorflow.Operations
             Func<LoopVar<TItem>, LoopVar<TItem>> body,
             LoopVar<TItem> original_loop_vars,
             Tensor[] loop_vars,
-            TensorShape[] shape_invariants) where TItem : IFromMergeVars<TItem>, new()
+            Shape[] shape_invariants) where TItem : IFromMergeVars<TItem>, new()
         {
             var flat_loop_vars = nest.flatten2(original_loop_vars)
                 .Select(x => (ITensorOrTensorArray)x)
@@ -459,8 +459,8 @@ namespace Tensorflow.Operations
             // dynamically from the forward inference. Getting the shape right
             // for the zeros is only needed for the base case when the loop exits
             // without running any iterations.
-            var shape = grad.TensorShape;
-            if (shape.is_fully_defined())
+            var shape = grad.shape;
+            if (shape.IsFullyDefined)
             {
                 if (outer_context != null)
                     outer_context.Enter();

src/TensorFlowNET.Core/Operations/Initializers/InitializerArgs.cs

@@ -3,11 +3,11 @@
     public class InitializerArgs
     {
         public string Name { get; set; }
-        public TensorShape Shape { get; set; }
+        public Shape Shape { get; set; }
         public TF_DataType DType { get; set; }
         public bool VerifyShape { get; set; }
 
-        public InitializerArgs(TensorShape shape,
+        public InitializerArgs(Shape shape,
             TF_DataType dtype = TF_DataType.DtInvalid,
             bool verify_shape = false,
             string name = null)

src/TensorFlowNET.Core/Operations/Initializers/Zeros.cs

@@ -18,10 +18,10 @@ namespace Tensorflow.Operations.Initializers
 {
     public class Zeros : IInitializer
     {
-        TensorShape shape;
+        Shape shape;
         TF_DataType dtype;
 
-        public Zeros(TensorShape shape = null, TF_DataType dtype = TF_DataType.TF_FLOAT)
+        public Zeros(Shape shape = null, TF_DataType dtype = TF_DataType.TF_FLOAT)
         {
             this.shape = shape;
             this.dtype = dtype;

src/TensorFlowNET.Core/Operations/Losses/losses_impl.py.cs

@@ -135,8 +135,8 @@ namespace Tensorflow
             if (weights > 0)
             {
                 var weights_tensor = ops.convert_to_tensor(weights);
-                var labels_rank = labels.TensorShape.ndim;
-                var weights_shape = weights_tensor.TensorShape;
+                var labels_rank = labels.shape.ndim;
+                var weights_shape = weights_tensor.shape;
                 var weights_rank = weights_shape.ndim;
 
                 if (labels_rank > -1 && weights_rank > -1)

src/TensorFlowNET.Core/Operations/NnOps/BasicLSTMCell.cs

@@ -46,7 +46,7 @@ namespace Tensorflow
                 _activation = tf.nn.tanh();
         }
 
-        protected override void build(TensorShape input_shape)
+        protected override void build(Shape input_shape)
         {
             var input_depth = input_shape.dims.Last();
             var h_depth = _num_units;

src/TensorFlowNET.Core/Operations/NnOps/BasicRNNCell.cs

@@ -50,7 +50,7 @@ namespace Tensorflow
                 _activation = activation;
         }
 
-        protected override void build(TensorShape inputs_shape)
+        protected override void build(Shape inputs_shape)
         {
             var input_depth = inputs_shape.dims[inputs_shape.ndim - 1];
 

src/TensorFlowNET.Core/Operations/NnOps/ConvolutionInternal.cs

@@ -38,8 +38,8 @@ namespace Tensorflow.Operations
 
         public Tensor Apply(Tensors input, IVariableV1 filters)
         {
-            var filters_rank = filters.shape.rank;
-            var inputs_rank = input.shape.rank;
+            var filters_rank = filters.shape.ndim;
+            var inputs_rank = input.shape.ndim;
             var num_spatial_dims = args.NumSpatialDims;
             if (args.Rank == 1)
             {

src/TensorFlowNET.Core/Operations/NnOps/LayerRNNCell.cs

@@ -49,7 +49,7 @@ namespace Tensorflow
             _keras_style = false;
         }
 
-        protected virtual void build(TensorShape inputs_shape)
+        protected virtual void build(Shape inputs_shape)
         {
 
         }

src/TensorFlowNET.Core/Operations/NnOps/RNNCell.cs

@@ -42,7 +42,7 @@ namespace Tensorflow
     /// This operation results in an output matrix with `self.output_size` columns.
     /// If `self.state_size` is an integer, this operation also results in a new
     /// state matrix with `self.state_size` columns.  If `self.state_size` is a
-    /// (possibly nested tuple of) TensorShape object(s), then it should return a
+    /// (possibly nested tuple of) Shape object(s), then it should return a
     /// matching structure of Tensors having shape `[batch_size].concatenate(s)`
     /// for each `s` in `self.batch_size`.
     /// </summary>
@@ -70,9 +70,9 @@ namespace Tensorflow
         public List<IVariableV1> trainable_weights => throw new NotImplementedException();
         public List<IVariableV1> non_trainable_weights => throw new NotImplementedException();
 
-        public TensorShape output_shape => throw new NotImplementedException();
+        public Shape output_shape => throw new NotImplementedException();
 
-        public TensorShape BatchInputShape => throw new NotImplementedException();
+        public Shape BatchInputShape => throw new NotImplementedException();
 
         public TF_DataType DType => throw new NotImplementedException();
         protected bool built = false;

src/TensorFlowNET.Core/Operations/NnOps/rnn.cs

@@ -118,14 +118,14 @@ namespace Tensorflow.Operations
                     VariableScope varscope = scope1;
                     // Obtain the first sequence of the input
                     var first_input = inputs[0];
-                    if (first_input.TensorShape.rank != 1)
+                    if (first_input.shape.ndim != 1)
                     {
-                        var input_shape = first_input.TensorShape.with_rank_at_least(2);
+                        var input_shape = first_input.shape.with_rank_at_least(2);
                         fixed_batch_size = input_shape.dims[0];
                         var flat_inputs = nest.flatten2(inputs);
                         foreach (var flat_input in flat_inputs)
                         {
-                            input_shape = flat_input.TensorShape.with_rank_at_least(2);
+                            input_shape = flat_input.shape.with_rank_at_least(2);
                             batch_size = tensor_shape.dimension_at_index(input_shape, 0);
                             var input_size = input_shape[new Slice(1)];
                             fixed_batch_size.merge_with(batch_size);
@@ -138,7 +138,7 @@ namespace Tensorflow.Operations
                         }
                     }
                     else
-                        fixed_batch_size = first_input.TensorShape.with_rank_at_least(1).dims[0];
+                        fixed_batch_size = first_input.shape.with_rank_at_least(1).dims[0];
 
                     if (tensor_shape.dimension_value(fixed_batch_size) >= 0)
                         batch_size = tensor_shape.dimension_value(fixed_batch_size);
@@ -243,7 +243,7 @@ namespace Tensorflow.Operations
             var input_shape = array_ops.shape(flat_input[0]);
             var time_steps = input_shape.slice(0);
             var batch_size = _best_effort_input_batch_size(flat_input);
-            var inputs_got_shape = flat_input.Select(input_ => input_.TensorShape.with_rank_at_least(3)).ToArray();
+            var inputs_got_shape = flat_input.Select(input_ => input_.shape.with_rank_at_least(3)).ToArray();
 
             var dims = inputs_got_shape[0].dims.Take(2).ToArray();
             var (const_time_steps, const_batch_size) = (dims[0], dims[1]);
@@ -292,7 +292,7 @@ namespace Tensorflow.Operations
             string base_name = null;
             tf_with(ops.name_scope("dynamic_rnn"), scope => base_name = scope);
 
-            Func<string, TensorShape, TF_DataType, TensorArray> _create_ta = (name, element_shape, dtype_) =>
+            Func<string, Shape, TF_DataType, TensorArray> _create_ta = (name, element_shape, dtype_) =>
             {
                 var ta = new TensorArray(dtype: dtype_,
                                         size: time_steps,
@@ -309,7 +309,7 @@ namespace Tensorflow.Operations
                 foreach (var (i, out_size) in enumerate(flat_output_size))
                 {
                     output_ta.Add(_create_ta($"output_{i}",
-                        new TensorShape(const_batch_size).concatenate(
+                        new Shape(const_batch_size).concatenate(
                             _maybe_tensor_shape_from_tensor(out_size)),
                         _infer_state_dtype(dtype, state)));
                 }
@@ -317,7 +317,7 @@ namespace Tensorflow.Operations
                 foreach (var (i, flat_input_i) in enumerate(flat_input))
                 {
                     input_ta.Add(_create_ta($"input_{i}",
-                        new TensorShape(flat_input_i.dims.Skip(1).ToArray()),
+                        new Shape(flat_input_i.dims.Skip(1).ToArray()),
                         flat_input_i.dtype));
                 }
 
@@ -350,7 +350,7 @@ namespace Tensorflow.Operations
                     input_t = input_ta.Select(ta => ta.read(time1)).ToArray();
                     // Restore some shape information
                     foreach (var (input_, shape) in zip(input_t, inputs_got_shape))
-                        input_.set_shape(shape[new Slice(1)]);
+                        input_.shape = shape[new Slice(1)];
                 }
                 else
                 {
@@ -397,17 +397,17 @@ namespace Tensorflow.Operations
             foreach (var (output, output_size) in zip(final_outputs, flat_output_size))
             {
                 var shape = rnn_cell_impl._concat(new int[] { (int)const_time_steps, (int)const_batch_size }, output_size, @static: true);
-                output.set_shape(shape);
+                output.shape = shape;
             }
 
             return (final_outputs[0], final_state);
         }
 
-        private static TensorShape _maybe_tensor_shape_from_tensor(Tensor shape)
-            => shape.TensorShape;
+        private static Shape _maybe_tensor_shape_from_tensor(Tensor shape)
+            => shape.shape;
 
-        private static TensorShape _maybe_tensor_shape_from_tensor(int shape)
-            => new TensorShape(shape);
+        private static Shape _maybe_tensor_shape_from_tensor(int shape)
+            => new Shape(shape);
 
         private static TF_DataType _infer_state_dtype(TF_DataType explicit_dtype, Tensor state)
         {
@@ -424,7 +424,7 @@ namespace Tensorflow.Operations
         /// <returns></returns>
         public static Tensor _transpose_batch_time(Tensor x)
         {
-            var x_static_shape = x.TensorShape;
+            var x_static_shape = x.shape;
             if (x_static_shape.ndim == 1)
                 return x;
 
@@ -436,12 +436,12 @@ namespace Tensorflow.Operations
             };
             var x_t = array_ops.transpose(x, array_ops.concat(con1, 0));
 
-            var dims = new [] { x_static_shape.dims[1], x_static_shape.dims[0] }
+            var dims = new long[] { x_static_shape.dims[1], x_static_shape.dims[0] }
                 .ToList();
             dims.AddRange(x_static_shape.dims.Skip(2));
-            var shape = new TensorShape(dims.ToArray());
+            var shape = new Shape(dims.ToArray());
 
-            x_t.set_shape(shape);
+            x_t.shape = shape;
 
             return x_t;
         }
@@ -455,7 +455,7 @@ namespace Tensorflow.Operations
         {
             foreach (var input_ in flat_input)
             {
-                var shape = input_.TensorShape;
+                var shape = input_.shape;
                 if (shape.ndim < 0)
                     continue;
                 if (shape.ndim < 2)

src/TensorFlowNET.Core/Operations/NnOps/rnn_cell_impl.cs

@@ -32,18 +32,18 @@ namespace Tensorflow.Operations
             else if (p.ndim != 1)
                 throw new ValueError($"prefix tensor must be either a scalar or vector, but saw tensor: {p}");
 
-            var s_tensor_shape = new TensorShape(suffix);
+            var s_tensor_shape = new Shape(suffix);
             var s_static = s_tensor_shape.ndim > -1 ?
                 s_tensor_shape.dims :
                 null;
-            var s = s_tensor_shape.is_fully_defined() ?
+            var s = s_tensor_shape.IsFullyDefined ?
                 constant_op.constant(s_tensor_shape.dims, dtype: dtypes.int32) :
                 null;
 
             if (@static)
             {
                 if (p_static is null) return null;
-                var shape = new TensorShape(p_static).concatenate(s_static);
+                var shape = new Shape(p_static).concatenate(s_static);
                 throw new NotImplementedException("RNNCell _concat");
             }
             else
@@ -54,24 +54,24 @@ namespace Tensorflow.Operations
             }
         }
 
-        public static TensorShape _concat(int[] prefix, int suffix, bool @static = false)
+        public static Shape _concat(int[] prefix, int suffix, bool @static = false)
         {
-            var p = new TensorShape(prefix);
+            var p = new Shape(prefix);
             var p_static = prefix;
-            var p_tensor = p.is_fully_defined() ? constant_op.constant(p.as_list(), dtype: dtypes.int32) : null;
+            var p_tensor = p.IsFullyDefined ? constant_op.constant(p, dtype: dtypes.int32) : null;
 
-            var s_tensor_shape = new TensorShape(suffix);
+            var s_tensor_shape = new Shape(suffix);
             var s_static = s_tensor_shape.ndim > -1 ?
                 s_tensor_shape.dims :
                 null;
-            var s_tensor = s_tensor_shape.is_fully_defined() ?
+            var s_tensor = s_tensor_shape.IsFullyDefined ?
                 constant_op.constant(s_tensor_shape.dims, dtype: dtypes.int32) :
                 null;
 
             if (@static)
             {
                 if (p_static is null) return null;
-                var shape = new TensorShape(p_static).concatenate(s_static);
+                var shape = new Shape(p_static).concatenate(s_static);
                 return shape;
             }
             else

src/TensorFlowNET.Core/Operations/OpDefLibrary.cs

@@ -345,7 +345,7 @@ namespace Tensorflow
             return ByteString.CopyFromUtf8(value ?? string.Empty);
         }
 
-        public TensorShapeProto _MakeShape(TensorShape shape, AttrDef attr_def)
+        public TensorShapeProto _MakeShape(Shape shape, AttrDef attr_def)
         {
             return shape.as_proto();
         }
@@ -400,7 +400,7 @@ namespace Tensorflow
                     if (value == null && attr_def.DefaultValue != null)
                         attr_value.Shape = attr_def.DefaultValue.Shape;
 
-                    if (value is TensorShape val1)
+                    if (value is Shape val1)
                         attr_value.Shape = val1.as_proto();
                     else if (value is long[] val2)
                         attr_value.Shape = tensor_util.as_shape(val2);
@@ -409,7 +409,7 @@ namespace Tensorflow
 
                     break;
                 case "list(shape)":
-                    attr_value.List.Shape.AddRange((value as TensorShape[]).Select(x => _MakeShape(x, attr_def)));
+                    attr_value.List.Shape.AddRange((value as Shape[]).Select(x => _MakeShape(x, attr_def)));
                     break;
                 default:
                     throw new TypeError($"SetAttrValue: can't not convert attr_def.Type '{attr_def.Type}' to protos.");

src/TensorFlowNET.Core/Operations/Queues/FIFOQueue.cs

@@ -22,7 +22,7 @@ namespace Tensorflow.Queues
     {
         public FIFOQueue(int capacity,
             TF_DataType[] dtypes,
-            TensorShape[] shapes,
+            Shape[] shapes,
             string[] names = null,
             string shared_name = null,
             string name = "fifo_queue")

src/TensorFlowNET.Core/Operations/Queues/PaddingFIFOQueue.cs

@@ -25,7 +25,7 @@ namespace Tensorflow.Queues
     {
         public PaddingFIFOQueue(int capacity,
             TF_DataType[] dtypes,
-            TensorShape[] shapes,
+            Shape[] shapes,
             string[] names = null,
             string shared_name = null,
             string name = "padding_fifo_queue")

src/TensorFlowNET.Core/Operations/Queues/PriorityQueue.cs

@@ -24,7 +24,7 @@ namespace Tensorflow.Queues
     {
         public PriorityQueue(int capacity,
             TF_DataType[] dtypes,
-            TensorShape[] shapes,
+            Shape[] shapes,
             string[] names = null,
             string shared_name = null,
             string name = "priority_queue")
@@ -44,7 +44,7 @@ namespace Tensorflow.Queues
             _dtypes = dtypes1.ToArray();
 
             var shapes1 = shapes.ToList();
-            shapes1.Insert(0, new TensorShape());
+            shapes1.Insert(0, Shape.Null);
             _shapes = shapes1.ToArray();
         }
 

src/TensorFlowNET.Core/Operations/Queues/QueueBase.cs

@@ -23,12 +23,12 @@ namespace Tensorflow.Queues
     public class QueueBase
     {
         protected TF_DataType[] _dtypes;
-        protected TensorShape[] _shapes;
+        protected Shape[] _shapes;
         protected string[] _names;
         protected Tensor _queue_ref;
         protected string _name;
 
-        public QueueBase(TF_DataType[] dtypes, TensorShape[] shapes, string[] names)
+        public QueueBase(TF_DataType[] dtypes, Shape[] shapes, string[] names)
         {
             _dtypes = dtypes;
             _shapes = shapes;

src/TensorFlowNET.Core/Operations/Queues/RandomShuffleQueue.cs

@@ -26,7 +26,7 @@ namespace Tensorflow.Queues
         public RandomShuffleQueue(int capacity,
             int min_after_dequeue,
             TF_DataType[] dtypes,
-            TensorShape[] shapes,
+            Shape[] shapes,
             string[] names = null,
             int? seed = null,
             string shared_name = null,

src/TensorFlowNET.Core/Operations/_GraphTensorArray.cs

@@ -37,7 +37,7 @@ namespace Tensorflow.Operations
         bool _infer_shape;
         public bool infer_shape => _infer_shape;
         public bool _dynamic_size;
-        public List<TensorShape> _element_shape;
+        public List<Shape> _element_shape;
 
         public List<Tensor> _colocate_with;
 
@@ -47,7 +47,7 @@ namespace Tensorflow.Operations
 
         public _GraphTensorArray(TF_DataType dtype, Tensor size, bool? dynamic_size = null,
             bool? clear_after_read = null, string tensor_array_name = null, Tensor handle = null, Tensor flow = null,
-            bool infer_shape = true, TensorShape element_shape = null,
+            bool infer_shape = true, Shape element_shape = null,
             bool colocate_with_first_write_call = true, string name = null)
         {
             clear_after_read = clear_after_read ?? true;
@@ -66,12 +66,12 @@ namespace Tensorflow.Operations
             if (element_shape == null)
             {
                 _infer_shape = infer_shape;
-                _element_shape = new List<TensorShape> { };
+                _element_shape = new List<Shape> { };
             }
             else
             {
                 _infer_shape = true;
-                _element_shape = new List<TensorShape> { element_shape };
+                _element_shape = new List<Shape> { element_shape };
             }
 
             tf_with(ops.name_scope(name, "TensorArray", new { handle, size, flow }), scope =>
@@ -124,7 +124,7 @@ namespace Tensorflow.Operations
                 value = ops.convert_to_tensor(value, preferred_dtype: _dtype, name: "value");
                 if (_infer_shape)
                 {
-                    var shape = new TensorShape(value.TensorShape.dims.Skip(1).ToArray());
+                    var shape = new Shape(value.shape.dims.Skip(1).ToArray());
                     _merge_element_shape(shape);
                 }
 
@@ -149,7 +149,7 @@ namespace Tensorflow.Operations
             });
         }
 
-        public void _merge_element_shape(TensorShape shape)
+        public void _merge_element_shape(Shape shape)
         {
             _element_shape.Add(shape);
         }
@@ -169,7 +169,7 @@ namespace Tensorflow.Operations
                 name: name);
 
             if (_element_shape != null)
-                value.set_shape(_element_shape[0].dims);
+                value.shape = _element_shape[0].dims;
 
             return value;
         }
@@ -207,7 +207,7 @@ namespace Tensorflow.Operations
 
         public Tensor gather(Tensor indices, string name = null)
         {
-            var element_shape = new TensorShape();
+            var element_shape = Shape.Null;
 
             if (_element_shape.Count > 0)
                 element_shape = _element_shape[0];

src/TensorFlowNET.Core/Operations/array_ops.cs

@@ -71,7 +71,7 @@ namespace Tensorflow
                 verify_shape: verify_shape,
                 allow_broadcast: false);
 
-        public static Tensor zeros(TensorShape shape, TF_DataType dtype = TF_DataType.TF_FLOAT, string name = null)
+        public static Tensor zeros(Shape shape, TF_DataType dtype = TF_DataType.TF_FLOAT, string name = null)
         {
             dtype = dtype.as_base_dtype();
 
@@ -130,9 +130,9 @@ namespace Tensorflow
                 var tensor_tensor = ops.convert_to_tensor(tensor, name: "tensor");
                 var mask_tensor = ops.convert_to_tensor(mask, name: "mask");
 
-                var shape_mask = mask_tensor.TensorShape;
+                var shape_mask = mask_tensor.shape;
                 var ndims_mask = shape_mask.ndim;
-                var shape_tensor = tensor_tensor.TensorShape;
+                var shape_tensor = tensor_tensor.shape;
 
                 if (ndims_mask < 1)
                     throw new ValueError("mask cannot be scalar.");
@@ -146,9 +146,9 @@ namespace Tensorflow
                 }, 0);
                 tensor_tensor = reshape(tensor_tensor, shape1);
                 var first_dim = shape_tensor.dims.Skip(axis).Take(ndims_mask).First();
-                var s1 = tensor_shape.as_shape(shape_tensor.dims.Take(axis).ToArray());
+                var s1 = new Shape(shape_tensor.dims.Take(axis).ToArray());
                 var s2 = s1.concatenate(new[] { first_dim }).concatenate(shape_tensor.dims.Skip(axis + ndims_mask).ToArray());
-                tensor_tensor.set_shape(s2);
+                tensor_tensor.shape = s2;
 
                 mask_tensor = reshape(mask_tensor, new[] { -1 });
                 return _apply_mask_1d(tensor_tensor, mask_tensor, axis);
@@ -186,10 +186,10 @@ namespace Tensorflow
 
         private static Tensor _constant_if_small(int value, Tensor shape)
         {
-            return shape < 1000UL;
+            return shape < 1000L;
         }
 
-        private static Tensor _constant_if_small<T>(T value, TensorShape shape, TF_DataType dtype, string name)
+        private static Tensor _constant_if_small<T>(T value, Shape shape, TF_DataType dtype, string name)
         {
             if (shape.size < 1000)
             {
@@ -364,7 +364,7 @@ namespace Tensorflow
                 tensor = ops.convert_to_tensor(tensor, name: "tensor");
 
                 // is_fully_defined return unexpected value.
-                if (optimize && tensor.shape.is_fully_defined() && dtype != TF_DataType.TF_VARIANT)
+                if (optimize && tensor.shape.IsFullyDefined && dtype != TF_DataType.TF_VARIANT)
                 {
 
                 }
@@ -384,7 +384,7 @@ namespace Tensorflow
         public static Tensor reshape(Tensor tensor, Tensor shape, string name = null)
             => gen_array_ops.reshape(tensor, shape, name: name);
 
-        public static Tensor reshape(Tensor tensor, TensorShape shape, string name = null)
+        public static Tensor reshape(Tensor tensor, Shape shape, string name = null)
             => gen_array_ops.reshape(tensor, shape, name: name);
 
         public static Tensor reshape(Tensor tensor, object[] shape, string name = null)
@@ -427,7 +427,7 @@ namespace Tensorflow
             });
         }
 
-        public static Tensor ones(TensorShape shape, TF_DataType dtype = TF_DataType.TF_FLOAT, string name = null)
+        public static Tensor ones(Shape shape, TF_DataType dtype = TF_DataType.TF_FLOAT, string name = null)
             => tf_with(ops.name_scope(name, "ones", shape), scope =>
             {
                 dtype = dtype.as_base_dtype();
@@ -502,7 +502,7 @@ namespace Tensorflow
                 return ops.convert_to_tensor(values, name: name);
             }
 
-            var value_shape = ops.convert_to_tensor(values[0], name: name).TensorShape;
+            var value_shape = ops.convert_to_tensor(values[0], name: name).shape;
 
             return gen_array_ops.pack(values, axis: axis, name: name);
         }
@@ -512,7 +512,7 @@ namespace Tensorflow
             if (num == null)
             {
                 value = ops.convert_to_tensor(value);
-                var value_shape = value.TensorShape;
+                var value_shape = value.shape;
                 num = (int)value_shape.dims[axis];
             }
 
@@ -588,8 +588,8 @@ namespace Tensorflow
 
                 if (!tf.Context.executing_eagerly())
                 {
-                    var input_shape = input.TensorShape;
-                    if (optimize && input.ndim > -1 && input_shape.is_fully_defined())
+                    var input_shape = input.shape;
+                    if (optimize && input.ndim > -1 && input_shape.IsFullyDefined)
                     {
                         var nd = np.array(input.shape.dims).astype(out_type.as_system_dtype());
                         return constant_op.constant(nd, name: name);
@@ -610,7 +610,7 @@ namespace Tensorflow
                 var input_shape = input_tensor.shape;
                 if (optimize)
                 {
-                    if (input_shape.is_fully_defined())
+                    if (input_shape.IsFullyDefined)
                     {
                         return constant_op.constant(input_shape.size, dtype: out_type, name: name);
                     }
@@ -633,7 +633,7 @@ namespace Tensorflow
                 tensor = ops.convert_to_tensor(tensor, name: "tensor");
 
                 // is_fully_defined return unexpected value.
-                if (optimize && tensor.shape.is_fully_defined() && dtype != TF_DataType.TF_VARIANT)
+                if (optimize && tensor.shape.IsFullyDefined && dtype != TF_DataType.TF_VARIANT)
                 {
 
                 }
@@ -906,7 +906,7 @@ namespace Tensorflow
             return gen_array_ops.gather_v2(@params, indices, axis, name: name);
         }
 
-        public static Tensor transpose<T1>(T1 a, TensorShape perm, string name = "transpose", bool conjugate = false)
+        public static Tensor transpose<T1>(T1 a, Shape perm, string name = "transpose", bool conjugate = false)
         {
             return tf_with(ops.name_scope(name, "transpose", new { a }), scope =>
             {
@@ -1005,9 +1005,9 @@ namespace Tensorflow
             if (!tf.Context.executing_eagerly())
             {
                 var paddings_constant = tensor_util.constant_value(paddings);
-                var input_shape = result.op.inputs[0].TensorShape;
+                var input_shape = result.op.inputs[0].shape;
                 if (input_shape.ndim > -1 &&
-                    !result.TensorShape.is_fully_defined() &&
+                    !result.shape.IsFullyDefined &&
                     !(paddings_constant is null))
                 {
                     var new_shape = new List<int>();
@@ -1018,14 +1018,14 @@ namespace Tensorflow
                         else
                             new_shape.Add(np.sum(padding) + dim);
                     }
-                    result.set_shape(new_shape.ToArray());
+                    result.shape = new_shape.ToArray();
                 }
             }
 
             return result;
         }
 
-        public static Tensor placeholder(TF_DataType dtype, TensorShape shape = null, string name = null)
+        public static Tensor placeholder(TF_DataType dtype, Shape shape = null, string name = null)
         {
             if (tf.Context.executing_eagerly())
                 throw new RuntimeError("tf.placeholder() is not compatible with eager execution.");

src/TensorFlowNET.Core/Operations/clip_ops.cs

@@ -29,9 +29,9 @@ namespace Tensorflow
                 var t_min = math_ops.minimum(values, clip_value_max);
                 // Assert that the shape is compatible with the initial shape,
                 // to prevent unintentional broadcasting.
-                _ = values.TensorShape.merge_with(t_min.shape);
+                _ = values.shape.merge_with(t_min.shape);
                 var t_max = math_ops.maximum(t_min, clip_value_min, name: name);
-                _ = values.TensorShape.merge_with(t_max.shape);
+                _ = values.shape.merge_with(t_max.shape);
 
                 return t_max;
             });

src/TensorFlowNET.Core/Operations/confusion_matrix.py.cs

@@ -38,9 +38,9 @@ namespace Tensorflow
             {
                 predictions = ops.convert_to_tensor(predictions);
                 labels = ops.convert_to_tensor(labels);
-                var predictions_shape = predictions.TensorShape;
+                var predictions_shape = predictions.shape;
                 var predictions_rank = predictions_shape.ndim;
-                var labels_shape = labels.TensorShape;
+                var labels_shape = labels.shape;
                 var labels_rank = labels_shape.ndim;
                 if (labels_rank > -1 && predictions_rank > -1)
                 {

src/TensorFlowNET.Core/Operations/control_flow_ops.cs

@@ -319,7 +319,7 @@ namespace Tensorflow
                 return gen_array_ops.identity(data, name: name);
         }
 
-        public static void _SetShapeInvariants(Tensor[] input_vars, Tensor[] enter_vars, TensorShape[] shapes = null)
+        public static void _SetShapeInvariants(Tensor[] input_vars, Tensor[] enter_vars, Shape[] shapes = null)
         {
             if (shapes == null)
                 return;
@@ -327,7 +327,7 @@ namespace Tensorflow
             var flat_shapes = nest.flatten2(shapes);
             foreach (var (inp, var, shape) in zip(input_vars, enter_vars, flat_shapes))
             {
-                var.set_shape(shape);
+                var.shape = shape;
             }
         }
 
@@ -706,7 +706,7 @@ namespace Tensorflow
         /// <param name="loop_vars"></param>
         /// <param name="shape_invariants"></param>
         public static TItem while_loop<TItem>(Func<TItem, Tensor> cond, Func<TItem, TItem> body, TItem loop_vars,
-            TensorShape[] shape_invariants = null,
+            Shape[] shape_invariants = null,
             int parallel_iterations = 10,
             bool back_prop = true,
             bool swap_memory = false,
@@ -803,7 +803,7 @@ namespace Tensorflow
                     data, frame_name, is_constant, parallel_iterations, name: name);
 
             if (use_input_shape)
-                result.set_shape(data.TensorShape);
+                result.shape = data.shape;
 
             return result;
         }

src/TensorFlowNET.Core/Operations/dataset_ops.cs

@@ -7,7 +7,7 @@ namespace Tensorflow
 {
     public class dataset_ops
     {
-        public Tensor tensor_dataset(Tensor[] components, TensorShape[] output_shapes, string name = null)
+        public Tensor tensor_dataset(Tensor[] components, Shape[] output_shapes, string name = null)
             => tf.Context.ExecuteOp("TensorDataset", name, new ExecuteOpArgs()
             {
                 OpInputArgs = new object[] { components }
@@ -20,29 +20,29 @@ namespace Tensorflow
         /// <param name="output_shapes"></param>
         /// <param name="name"></param>
         /// <returns></returns>
-        public Tensor tensor_slice_dataset(Tensor[] components, TensorShape[] output_shapes, string name = null)
+        public Tensor tensor_slice_dataset(Tensor[] components, Shape[] output_shapes, string name = null)
             => tf.Context.ExecuteOp("TensorSliceDataset", name, new ExecuteOpArgs()
             {
                 OpInputArgs = new object[] { components }
             }.SetAttributes(new { output_shapes }));
 
-        public Tensor range_dataset(Tensor start, Tensor stop, Tensor step, TF_DataType[] output_types, TensorShape[] output_shapes, string name = null)
+        public Tensor range_dataset(Tensor start, Tensor stop, Tensor step, TF_DataType[] output_types, Shape[] output_shapes, string name = null)
             => tf.Context.ExecuteOp("RangeDataset", name, new ExecuteOpArgs(start, stop, step)
                 .SetAttributes(new { output_types, output_shapes }));
 
-        public Tensor repeat_dataset(Tensor input_dataset, Tensor count, TF_DataType[] output_types, TensorShape[] output_shapes, string name = null)
+        public Tensor repeat_dataset(Tensor input_dataset, Tensor count, TF_DataType[] output_types, Shape[] output_shapes, string name = null)
             => tf.Context.ExecuteOp("RepeatDataset", name, new ExecuteOpArgs(input_dataset, count)
                 .SetAttributes(new { output_types, output_shapes }));
 
         public Tensor shard_dataset(Tensor input_dataset, Tensor num_shards, Tensor index,
-            TF_DataType[] output_types, TensorShape[] output_shapes,
+            TF_DataType[] output_types, Shape[] output_shapes,
             bool require_non_empty = false, string name = null)
                 => tf.Context.ExecuteOp("ShardDataset", name, new ExecuteOpArgs(input_dataset, num_shards, index)
                     .SetAttributes(new { require_non_empty, output_types, output_shapes }));
 
         public Tensor zip_dataset(Tensor[] input_datasets,
             TF_DataType[] output_types,
-            TensorShape[] output_shapes,
+            Shape[] output_shapes,
             string name = null)
                 => tf.Context.ExecuteOp("ZipDataset", name, new ExecuteOpArgs()
                 {
@@ -51,14 +51,14 @@ namespace Tensorflow
 
         public Tensor shuffle_dataset_v3(Tensor input_dataset, Tensor buffer_size,
             Tensor seed, Tensor seed2, Tensor seed_generator,
-            TF_DataType[] output_types, TensorShape[] output_shapes,
+            TF_DataType[] output_types, Shape[] output_shapes,
             bool reshuffle_each_iteration = true,
             string name = null)
                 => tf.Context.ExecuteOp("ShuffleDatasetV3", name, new ExecuteOpArgs(input_dataset, buffer_size, seed, seed2, seed_generator)
                     .SetAttributes(new { reshuffle_each_iteration, output_types, output_shapes }));
 
         public Tensor skip_dataset(Tensor input_dataset, Tensor count,
-            TF_DataType[] output_types, TensorShape[] output_shapes,
+            TF_DataType[] output_types, Shape[] output_shapes,
             string name = null)
                 => tf.Context.ExecuteOp("SkipDataset", name, new ExecuteOpArgs(input_dataset, count)
                     .SetAttributes(new { output_types, output_shapes }));
@@ -67,13 +67,13 @@ namespace Tensorflow
             => tf.Context.ExecuteOp("DummySeedGenerator", name, new ExecuteOpArgs());
 
         public Tensor concatenate_dataset(Tensor input_dataset, Tensor another_dataset,
-            TF_DataType[] output_types, TensorShape[] output_shapes,
+            TF_DataType[] output_types, Shape[] output_shapes,
             string name = null)
                 => tf.Context.ExecuteOp("ConcatenateDataset", name, new ExecuteOpArgs(input_dataset, another_dataset)
                     .SetAttributes(new { output_types, output_shapes }));
 
         public Tensor cache_dataset_v2(Tensor input_dataset, Tensor filename, Tensor cache,
-            TF_DataType[] output_types, TensorShape[] output_shapes,
+            TF_DataType[] output_types, Shape[] output_shapes,
             string name = null)
                 => tf.Context.ExecuteOp("CacheDatasetV2", name, new ExecuteOpArgs(input_dataset, filename, cache)
                     .SetAttributes(new { output_types, output_shapes }));
@@ -91,7 +91,7 @@ namespace Tensorflow
         /// <returns></returns>
         public Tensor batch_dataset_v2(Tensor input_dataset, Tensor buffer_size,
             Tensor drop_remainder,
-            TF_DataType[] output_types, TensorShape[] output_shapes,
+            TF_DataType[] output_types, Shape[] output_shapes,
             bool parallel_copy = false,
             string name = null)
                 => tf.Context.ExecuteOp("BatchDatasetV2", name, 
@@ -118,7 +118,7 @@ namespace Tensorflow
         /// <param name="name"></param>
         /// <returns></returns>
         public Tensor prefetch_dataset(Tensor input_dataset, Tensor buffer_size,
-            TF_DataType[] output_types, TensorShape[] output_shapes,
+            TF_DataType[] output_types, Shape[] output_shapes,
             int? slack_period = 0,
             bool legacy_autotune = true,
             string name = null)
@@ -141,7 +141,7 @@ namespace Tensorflow
         /// <param name="name"></param>
         /// <returns></returns>
         public Tensor take_dataset(Tensor input_dataset, Tensor count,
-            TF_DataType[] output_types, TensorShape[] output_shapes,
+            TF_DataType[] output_types, Shape[] output_shapes,
             string name = null)
                 => tf.Context.ExecuteOp("TakeDataset", name, new ExecuteOpArgs(input_dataset, count)
                     .SetAttributes(new { output_types, output_shapes }));
@@ -157,7 +157,7 @@ namespace Tensorflow
         /// <param name="name"></param>
         /// <returns></returns>
         public Tensor optimize_dataset(Tensor input_dataset, Tensor optimizations,
-            TF_DataType[] output_types, TensorShape[] output_shapes,
+            TF_DataType[] output_types, Shape[] output_shapes,
             string[] optimization_configs = null,
             string name = null)
                 => tf.Context.ExecuteOp("OptimizeDataset", name, new ExecuteOpArgs(input_dataset, optimizations)
@@ -170,7 +170,7 @@ namespace Tensorflow
 
         public Tensor optimize_dataset_v2(Tensor input_dataset, Tensor optimizations_enabled,
             Tensor optimizations_disabled, Tensor optimizations_default,
-            TF_DataType[] output_types, TensorShape[] output_shapes,
+            TF_DataType[] output_types, Shape[] output_shapes,
             string[] optimization_configs = null,
             string name = null)
                 => tf.Context.ExecuteOp("OptimizeDatasetV2", name, new ExecuteOpArgs(input_dataset, 
@@ -193,7 +193,7 @@ namespace Tensorflow
         /// <param name="name"></param>
         /// <returns></returns>
         public Tensor model_dataset(Tensor input_dataset,
-            TF_DataType[] output_types, TensorShape[] output_shapes,
+            TF_DataType[] output_types, Shape[] output_shapes,
             AutotuneAlgorithm algorithm, long cpu_budget, long ram_budget,
             string name = null)
                 => tf.Context.ExecuteOp("ModelDataset", name, new ExecuteOpArgs(input_dataset)
@@ -213,7 +213,7 @@ namespace Tensorflow
         /// <param name="output_shapes"></param>
         /// <param name="name"></param>
         /// <returns>A tuple of `Tensor` objects (handle, deleter).</returns>
-        public (Tensor, Tensor) anonymous_iterator_v2(TF_DataType[] output_types, TensorShape[] output_shapes, string name = null)
+        public (Tensor, Tensor) anonymous_iterator_v2(TF_DataType[] output_types, Shape[] output_shapes, string name = null)
         {
             var results = tf.Context.ExecuteOp("AnonymousIteratorV2", name, 
                 new ExecuteOpArgs().SetAttributes(new { output_types, output_shapes }));
@@ -237,7 +237,7 @@ namespace Tensorflow
         /// <param name="iterator"></param>
         /// <param name="name"></param>
         /// <returns></returns>
-        public Tensor map_dataset(Tensor dataset, ConcreteFunction f, TF_DataType[] output_types, TensorShape[] output_shapes,
+        public Tensor map_dataset(Tensor dataset, ConcreteFunction f, TF_DataType[] output_types, Shape[] output_shapes,
             bool use_inter_op_parallelism = true, bool preserve_cardinality = false, string name = null)
                 => tf.Context.ExecuteOp("MapDataset", name, new ExecuteOpArgs(dataset, new Tensor[0])
                     .SetAttributes(new
@@ -258,7 +258,7 @@ namespace Tensorflow
         /// <param name="output_shapes"></param>
         /// <param name="name"></param>
         /// <returns></returns>
-        public Tensor filter_dataset(Tensor dataset, ConcreteFunction predicate, TF_DataType[] output_types, TensorShape[] output_shapes,
+        public Tensor filter_dataset(Tensor dataset, ConcreteFunction predicate, TF_DataType[] output_types, Shape[] output_shapes,
             string name = null)
                 => tf.Context.ExecuteOp("FilterDataset", name, new ExecuteOpArgs(dataset, new Tensor[0])
                     .SetAttributes(new
@@ -277,7 +277,7 @@ namespace Tensorflow
         /// <param name="output_shapes"></param>
         /// <param name="name"></param>
         /// <returns></returns>
-        public Tensor flat_map_dataset(Tensor dataset, ConcreteFunction f, TF_DataType[] output_types, TensorShape[] output_shapes,
+        public Tensor flat_map_dataset(Tensor dataset, ConcreteFunction f, TF_DataType[] output_types, Shape[] output_shapes,
             string name = null)
                 => tf.Context.ExecuteOp("FlatMapDataset", name, new ExecuteOpArgs(dataset, new Tensor[0])
                     .SetAttributes(new { f, output_types, output_shapes }));
@@ -294,7 +294,7 @@ namespace Tensorflow
         /// <param name="name"></param>
         /// <returns></returns>
         public Tensor parallel_map_dataset_v2(Tensor dataset, Tensor num_parallel_calls, ConcreteFunction f,
-            TF_DataType[] output_types, TensorShape[] output_shapes,
+            TF_DataType[] output_types, Shape[] output_shapes,
             bool use_inter_op_parallelism = true,
             string deterministic = "default",
             bool preserve_cardinality = false,
@@ -329,7 +329,7 @@ namespace Tensorflow
         /// <param name="output_shapes"></param>
         /// <param name="name"></param>
         /// <returns></returns>
-        public Tensor[] iterator_get_next(Tensor iterator, TF_DataType[] output_types, TensorShape[] output_shapes, string name = null)
+        public Tensor[] iterator_get_next(Tensor iterator, TF_DataType[] output_types, Shape[] output_shapes, string name = null)
             => tf.Context.ExecuteOp("IteratorGetNext", name, new ExecuteOpArgs(iterator)
                 .SetAttributes(new { output_types, output_shapes }));
     }

src/TensorFlowNET.Core/Operations/functional_ops.cs

@@ -169,16 +169,16 @@ namespace Tensorflow
 
                 var results_flat = bodyItem.Accs_ta.Select(r => r.stack()).ToArray();
 
-                var n_static = new Dimension(tensor_shape.dimension_value(elems_flat[0].TensorShape.with_rank_at_least(1).dims[0]));
+                var n_static = new Dimension(tensor_shape.dimension_value(elems_flat[0].shape.with_rank_at_least(1).dims[0]));
 
                 foreach (var elem in elems_flat.Skip(1))
                 {
-                    n_static.merge_with(new Dimension(tensor_shape.dimension_value(elem.TensorShape.with_rank_at_least(1).dims[0])));
+                    n_static.merge_with(new Dimension(tensor_shape.dimension_value(elem.shape.with_rank_at_least(1).dims[0])));
                 }
 
                 foreach (Tensor r in results_flat)
                 {
-                    r.set_shape(new TensorShape(n_static).concatenate(r.dims.Skip(1).ToArray()));
+                    r.shape = new Shape(n_static).concatenate(r.dims.Skip(1).ToArray());
                 }
 
                 // todo get working when the above caching_device is fixed

src/TensorFlowNET.Core/Operations/gen_array_ops.cs

@@ -482,11 +482,7 @@ namespace Tensorflow
         /// <param name="name"> A name for the operation (optional).</param>
         /// <returns> `Tensor`. Has the same type as `s0`.</returns>
         public static Tensor broadcast_args(Tensor s0, Tensor s1, string name = null)
-        {
-            var _op = tf.OpDefLib._apply_op_helper("BroadcastArgs", name, args: new { s0, s1, name });
-
-            return _op.outputs[0];
-        }
+            => tf.Context.ExecuteOp("BroadcastArgs", name, new ExecuteOpArgs(s0, s1));
 
         /// <summary>
         /// Broadcast an array for a compatible shape.

src/TensorFlowNET.Core/Operations/gen_data_flow_ops.cs

@@ -41,7 +41,7 @@ namespace Tensorflow
         }
 
         public static (Tensor, Tensor) tensor_array_v3<T>(T size, TF_DataType dtype = TF_DataType.DtInvalid,
-            TensorShape element_shape = null, bool dynamic_size = false, bool clear_after_read = true,
+            Shape element_shape = null, bool dynamic_size = false, bool clear_after_read = true,
             bool identical_element_shapes = false, string tensor_array_name = "", string name = null)
         {
             var _op = tf.OpDefLib._apply_op_helper("TensorArrayV3", name, new
@@ -72,7 +72,7 @@ namespace Tensorflow
             return _op.output;
         }
 
-        public static Tensor padding_fifo_queue_v2(TF_DataType[] component_types, TensorShape[] shapes,
+        public static Tensor padding_fifo_queue_v2(TF_DataType[] component_types, Shape[] shapes,
             int capacity = -1, string container = "", string shared_name = "",
             string name = null)
         {
@@ -88,7 +88,7 @@ namespace Tensorflow
             return _op.output;
         }
 
-        public static Tensor fifo_queue_v2(TF_DataType[] component_types, TensorShape[] shapes,
+        public static Tensor fifo_queue_v2(TF_DataType[] component_types, Shape[] shapes,
             int capacity = -1, string container = "", string shared_name = "",
             string name = null)
         {
@@ -104,7 +104,7 @@ namespace Tensorflow
             return _op.output;
         }
 
-        public static Tensor priority_queue_v2(TF_DataType[] component_types, TensorShape[] shapes,
+        public static Tensor priority_queue_v2(TF_DataType[] component_types, Shape[] shapes,
             int capacity = -1, string container = "", string shared_name = "",
             string name = null)
         {
@@ -120,7 +120,7 @@ namespace Tensorflow
             return _op.output;
         }
 
-        public static Tensor random_shuffle_queue_v2(TF_DataType[] component_types, TensorShape[] shapes,
+        public static Tensor random_shuffle_queue_v2(TF_DataType[] component_types, Shape[] shapes,
             int capacity = -1, int min_after_dequeue = 0, int seed = 0, int seed2 = 0,
             string container = "", string shared_name = "", string name = null)
         {
@@ -259,7 +259,7 @@ namespace Tensorflow
         }
 
         public static Tensor tensor_array_gather_v3(Tensor handle, Tensor indices, Tensor flow_in,
-            TF_DataType dtype, TensorShape element_shape = null, string name = null)
+            TF_DataType dtype, Shape element_shape = null, string name = null)
         {
             var _op = tf.OpDefLib._apply_op_helper("TensorArrayGatherV3", name, new
             {

src/TensorFlowNET.Core/Operations/gen_math_ops.cs

@@ -397,8 +397,12 @@ namespace Tensorflow
         /// <param name="y"></param>
         /// <param name="name"></param>
         /// <returns></returns>
-        public static Tensor equal<Tx, Ty>(Tx x, Ty y, string name = null)
-            => tf.Context.ExecuteOp("Equal", name, new ExecuteOpArgs(x, y));
+        public static Tensor equal<Tx, Ty>(Tx x, Ty y, bool incompatible_shape_error = true, string name = null)
+            => tf.Context.ExecuteOp("Equal", name, new ExecuteOpArgs(x, y)
+                .SetAttributes(new
+                {
+                    incompatible_shape_error
+                }));
 
         /// <summary>
         /// Returns the truth value of (x != y) element-wise.

src/TensorFlowNET.Core/Operations/gen_ops.cs

@@ -88,7 +88,7 @@ namespace Tensorflow.Operations
         ///    
         ///    Returns a <c>Tensor</c> of same shape and type as the elements of <c>inputs</c>.
         /// </remarks>
-        public static Tensor accumulate_n_v2(Tensor[] inputs, TensorShape shape, string name = "AccumulateNV2")
+        public static Tensor accumulate_n_v2(Tensor[] inputs, Shape shape, string name = "AccumulateNV2")
         {
             var dict = new Dictionary<string, object>();
             dict["inputs"] = inputs;
@@ -754,7 +754,7 @@ namespace Tensorflow.Operations
         ///    container.
         ///    The Operation can be fetched from the resulting Tensor, by fetching the Operation property from the result.
         /// </returns>
-        public static Tensor anonymous_iterator(TF_DataType[] output_types, TensorShape[] output_shapes, string name = "AnonymousIterator")
+        public static Tensor anonymous_iterator(TF_DataType[] output_types, Shape[] output_shapes, string name = "AnonymousIterator")
         {
             var dict = new Dictionary<string, object>();
             dict["output_types"] = output_types;
@@ -2559,7 +2559,7 @@ namespace Tensorflow.Operations
         ///    incomplete element has some undefined components in its value tuple,
         ///    and may be updated using BarrierInsertMany.
         /// </remarks>
-        public static Tensor barrier(TF_DataType[] component_types, TensorShape[] shapes = null, int? capacity = null, string container = null, string shared_name = null, string name = "Barrier")
+        public static Tensor barrier(TF_DataType[] component_types, Shape[] shapes = null, int? capacity = null, string container = null, string shared_name = null, string name = "Barrier")
         {
             var dict = new Dictionary<string, object>();
             dict["component_types"] = component_types;
@@ -2883,7 +2883,7 @@ namespace Tensorflow.Operations
         /// <returns>
         ///    The Operation can be fetched from the resulting Tensor, by fetching the Operation property from the result.
         /// </returns>
-        public static Tensor batch_dataset(Tensor input_dataset, Tensor batch_size, TF_DataType[] output_types, TensorShape[] output_shapes, string name = "BatchDataset")
+        public static Tensor batch_dataset(Tensor input_dataset, Tensor batch_size, TF_DataType[] output_types, Shape[] output_shapes, string name = "BatchDataset")
         {
             var dict = new Dictionary<string, object>();
             dict["input_dataset"] = input_dataset;
@@ -2918,7 +2918,7 @@ namespace Tensorflow.Operations
         /// <returns>
         ///    The Operation can be fetched from the resulting Tensor, by fetching the Operation property from the result.
         /// </returns>
-        public static Tensor batch_dataset_v2(Tensor input_dataset, Tensor batch_size, Tensor drop_remainder, TF_DataType[] output_types, TensorShape[] output_shapes, string name = "BatchDatasetV2")
+        public static Tensor batch_dataset_v2(Tensor input_dataset, Tensor batch_size, Tensor drop_remainder, TF_DataType[] output_types, Shape[] output_shapes, string name = "BatchDatasetV2")
         {
             var dict = new Dictionary<string, object>();
             dict["input_dataset"] = input_dataset;
@@ -4367,7 +4367,7 @@ namespace Tensorflow.Operations
         /// <returns>
         ///    The Operation can be fetched from the resulting Tensor, by fetching the Operation property from the result.
         /// </returns>
-        public static Tensor bytes_produced_stats_dataset(Tensor input_dataset, Tensor tag, TF_DataType[] output_types, TensorShape[] output_shapes, string name = "BytesProducedStatsDataset")
+        public static Tensor bytes_produced_stats_dataset(Tensor input_dataset, Tensor tag, TF_DataType[] output_types, Shape[] output_shapes, string name = "BytesProducedStatsDataset")
         {
             var dict = new Dictionary<string, object>();
             dict["input_dataset"] = input_dataset;
@@ -4587,7 +4587,7 @@ namespace Tensorflow.Operations
         ///    (e.g. cannot be opened, contains tensors of the wrong shape / size), an error
         ///    will the returned when used.
         /// </remarks>
-        public static Tensor cache_dataset(Tensor input_dataset, Tensor filename, TF_DataType[] output_types, TensorShape[] output_shapes, string name = "CacheDataset")
+        public static Tensor cache_dataset(Tensor input_dataset, Tensor filename, TF_DataType[] output_types, Shape[] output_shapes, string name = "CacheDataset")
         {
             var dict = new Dictionary<string, object>();
             dict["input_dataset"] = input_dataset;
@@ -4802,7 +4802,7 @@ namespace Tensorflow.Operations
         /// <returns>
         ///    The Operation can be fetched from the resulting Tensor, by fetching the Operation property from the result.
         /// </returns>
-        public static Tensor collective_bcast_recv(TF_DataType T, int group_size, int group_key, int instance_key, TensorShape shape, string name = "CollectiveBcastRecv")
+        public static Tensor collective_bcast_recv(TF_DataType T, int group_size, int group_key, int instance_key, Shape shape, string name = "CollectiveBcastRecv")
         {
             var dict = new Dictionary<string, object>();
             dict["T"] = T;
@@ -4837,7 +4837,7 @@ namespace Tensorflow.Operations
         /// <returns>
         ///    The Operation can be fetched from the resulting Tensor, by fetching the Operation property from the result.
         /// </returns>
-        public static Tensor collective_bcast_send(Tensor input, int group_size, int group_key, int instance_key, TensorShape shape, string name = "CollectiveBcastSend")
+        public static Tensor collective_bcast_send(Tensor input, int group_size, int group_key, int instance_key, Shape shape, string name = "CollectiveBcastSend")
         {
             var dict = new Dictionary<string, object>();
             dict["input"] = input;
@@ -5187,7 +5187,7 @@ namespace Tensorflow.Operations
         /// <returns>
         ///    The Operation can be fetched from the resulting Tensor, by fetching the Operation property from the result.
         /// </returns>
-        public static Tensor concatenate_dataset(Tensor input_dataset, Tensor another_dataset, TF_DataType[] output_types, TensorShape[] output_shapes, string name = "ConcatenateDataset")
+        public static Tensor concatenate_dataset(Tensor input_dataset, Tensor another_dataset, TF_DataType[] output_types, Shape[] output_shapes, string name = "ConcatenateDataset")
         {
             var dict = new Dictionary<string, object>();
             dict["input_dataset"] = input_dataset;
@@ -5232,7 +5232,7 @@ namespace Tensorflow.Operations
         ///    resets the aggregate to 0, and increments the global_step recorded by
         ///    the accumulator.
         /// </remarks>
-        public static Tensor conditional_accumulator(TF_DataType dtype, TensorShape shape, string container = null, string shared_name = null, string name = "ConditionalAccumulator")
+        public static Tensor conditional_accumulator(TF_DataType dtype, Shape shape, string container = null, string shared_name = null, string name = "ConditionalAccumulator")
         {
             var dict = new Dictionary<string, object>();
             dict["dtype"] = dtype;
@@ -7271,7 +7271,7 @@ namespace Tensorflow.Operations
         ///    The components of the single element of <c>input</c>.
         ///    The Operation can be fetched from the resulting Tensor, by fetching the Operation property from the result.
         /// </returns>
-        public static Tensor[] dataset_to_single_element(Tensor dataset, TF_DataType[] output_types, TensorShape[] output_shapes, string name = "DatasetToSingleElement")
+        public static Tensor[] dataset_to_single_element(Tensor dataset, TF_DataType[] output_types, Shape[] output_shapes, string name = "DatasetToSingleElement")
         {
             var dict = new Dictionary<string, object>();
             dict["dataset"] = dataset;
@@ -8294,7 +8294,7 @@ namespace Tensorflow.Operations
         /// <returns>
         ///    The Operation can be fetched from the resulting Tensor, by fetching the Operation property from the result.
         /// </returns>
-        public static Tensor dense_to_sparse_batch_dataset(Tensor input_dataset, Tensor batch_size, Tensor row_shape, TF_DataType[] output_types, TensorShape[] output_shapes, string name = "DenseToSparseBatchDataset")
+        public static Tensor dense_to_sparse_batch_dataset(Tensor input_dataset, Tensor batch_size, Tensor row_shape, TF_DataType[] output_types, Shape[] output_shapes, string name = "DenseToSparseBatchDataset")
         {
             var dict = new Dictionary<string, object>();
             dict["input_dataset"] = input_dataset;
@@ -10059,7 +10059,7 @@ namespace Tensorflow.Operations
         ///    Raises an error if the input tensor's shape does not match the specified shape.
         ///    Returns the input tensor otherwise.
         /// </remarks>
-        public static Tensor ensure_shape(Tensor input, TensorShape shape, string name = "EnsureShape")
+        public static Tensor ensure_shape(Tensor input, Shape shape, string name = "EnsureShape")
         {
             var dict = new Dictionary<string, object>();
             dict["input"] = input;
@@ -10584,7 +10584,7 @@ namespace Tensorflow.Operations
         ///    The handle to the queue.
         ///    The Operation can be fetched from the resulting Tensor, by fetching the Operation property from the result.
         /// </returns>
-        public static Tensor f_i_f_o_queue(TF_DataType[] component_types, TensorShape[] shapes = null, int? capacity = null, string container = null, string shared_name = null, string name = "FIFOQueue")
+        public static Tensor f_i_f_o_queue(TF_DataType[] component_types, Shape[] shapes = null, int? capacity = null, string container = null, string shared_name = null, string name = "FIFOQueue")
         {
             var dict = new Dictionary<string, object>();
             dict["component_types"] = component_types;
@@ -10632,7 +10632,7 @@ namespace Tensorflow.Operations
         ///    The handle to the queue.
         ///    The Operation can be fetched from the resulting Tensor, by fetching the Operation property from the result.
         /// </returns>
-        public static Tensor f_i_f_o_queue_v2(TF_DataType[] component_types, TensorShape[] shapes = null, int? capacity = null, string container = null, string shared_name = null, string name = "FIFOQueueV2")
+        public static Tensor f_i_f_o_queue_v2(TF_DataType[] component_types, Shape[] shapes = null, int? capacity = null, string container = null, string shared_name = null, string name = "FIFOQueueV2")
         {
             var dict = new Dictionary<string, object>();
             dict["component_types"] = component_types;
@@ -10670,7 +10670,7 @@ namespace Tensorflow.Operations
         ///    \"Fake\" output value. This should not be consumed by another op.
         ///    The Operation can be fetched from the resulting Tensor, by fetching the Operation property from the result.
         /// </returns>
-        public static Tensor fake_param(TF_DataType dtype, TensorShape shape, string name = "FakeParam")
+        public static Tensor fake_param(TF_DataType dtype, Shape shape, string name = "FakeParam")
         {
             var dict = new Dictionary<string, object>();
             dict["dtype"] = dtype;
@@ -11048,7 +11048,7 @@ namespace Tensorflow.Operations
         /// <returns>
         ///    The Operation can be fetched from the resulting Tensor, by fetching the Operation property from the result.
         /// </returns>
-        public static Tensor filter_by_last_component_dataset(Tensor input_dataset, TF_DataType[] output_types, TensorShape[] output_shapes, string name = "FilterByLastComponentDataset")
+        public static Tensor filter_by_last_component_dataset(Tensor input_dataset, TF_DataType[] output_types, Shape[] output_shapes, string name = "FilterByLastComponentDataset")
         {
             var dict = new Dictionary<string, object>();
             dict["input_dataset"] = input_dataset;
@@ -13428,7 +13428,7 @@ namespace Tensorflow.Operations
         /// <remarks>
         ///    The current implementation memmaps the tensor from a file.
         /// </remarks>
-        public static Tensor immutable_const(TF_DataType dtype, TensorShape shape, string memory_region_name, string name = "ImmutableConst")
+        public static Tensor immutable_const(TF_DataType dtype, Shape shape, string memory_region_name, string name = "ImmutableConst")
         {
             var dict = new Dictionary<string, object>();
             dict["dtype"] = dtype;
@@ -13547,7 +13547,7 @@ namespace Tensorflow.Operations
         ///    A tensor that will be provided using the infeed mechanism.
         ///    The Operation can be fetched from the resulting Tensor, by fetching the Operation property from the result.
         /// </returns>
-        public static Tensor infeed_dequeue(TF_DataType dtype, TensorShape shape, string name = "InfeedDequeue")
+        public static Tensor infeed_dequeue(TF_DataType dtype, Shape shape, string name = "InfeedDequeue")
         {
             var dict = new Dictionary<string, object>();
             dict["dtype"] = dtype;
@@ -13577,7 +13577,7 @@ namespace Tensorflow.Operations
         /// <remarks>
         ///    simultaneously as an XLA tuple.
         /// </remarks>
-        public static Tensor[] infeed_dequeue_tuple(TF_DataType[] dtypes, TensorShape[] shapes, string name = "InfeedDequeueTuple")
+        public static Tensor[] infeed_dequeue_tuple(TF_DataType[] dtypes, Shape[] shapes, string name = "InfeedDequeueTuple")
         {
             var dict = new Dictionary<string, object>();
             dict["dtypes"] = dtypes;
@@ -13608,7 +13608,7 @@ namespace Tensorflow.Operations
         /// <returns>
         ///    Returns the description of the operation
         /// </returns>
-        public static Operation infeed_enqueue(Tensor input, TensorShape shape = null, int? device_ordinal = null, string name = "InfeedEnqueue")
+        public static Operation infeed_enqueue(Tensor input, Shape shape = null, int? device_ordinal = null, string name = "InfeedEnqueue")
         {
             var dict = new Dictionary<string, object>();
             dict["input"] = input;
@@ -13641,7 +13641,7 @@ namespace Tensorflow.Operations
         /// <returns>
         ///    Returns the description of the operation
         /// </returns>
-        public static Operation infeed_enqueue_tuple(Tensor[] inputs, TensorShape[] shapes, int? device_ordinal = null, string name = "InfeedEnqueueTuple")
+        public static Operation infeed_enqueue_tuple(Tensor[] inputs, Shape[] shapes, int? device_ordinal = null, string name = "InfeedEnqueueTuple")
         {
             var dict = new Dictionary<string, object>();
             dict["inputs"] = inputs;
@@ -14163,7 +14163,7 @@ namespace Tensorflow.Operations
         ///    or "IteratorGetNext" op.
         ///    The Operation can be fetched from the resulting Tensor, by fetching the Operation property from the result.
         /// </returns>
-        public static Tensor iterator(string shared_name, string container, TF_DataType[] output_types, TensorShape[] output_shapes, string name = "Iterator")
+        public static Tensor iterator(string shared_name, string container, TF_DataType[] output_types, Shape[] output_shapes, string name = "Iterator")
         {
             var dict = new Dictionary<string, object>();
             dict["shared_name"] = shared_name;
@@ -14195,7 +14195,7 @@ namespace Tensorflow.Operations
         ///    A handle to an iterator resource.
         ///    The Operation can be fetched from the resulting Tensor, by fetching the Operation property from the result.
         /// </returns>
-        public static Tensor iterator_from_string_handle(Tensor string_handle, TF_DataType[] output_types = null, TensorShape[] output_shapes = null, string name = "IteratorFromStringHandle")
+        public static Tensor iterator_from_string_handle(Tensor string_handle, TF_DataType[] output_types = null, Shape[] output_shapes = null, string name = "IteratorFromStringHandle")
         {
             var dict = new Dictionary<string, object>();
             dict["string_handle"] = string_handle;
@@ -14224,7 +14224,7 @@ namespace Tensorflow.Operations
         /// <returns>
         ///    The Operation can be fetched from the resulting Tensor, by fetching the Operation property from the result.
         /// </returns>
-        public static Tensor[] iterator_get_next(Tensor iterator, TF_DataType[] output_types, TensorShape[] output_shapes, string name = "IteratorGetNext")
+        public static Tensor[] iterator_get_next(Tensor iterator, TF_DataType[] output_types, Shape[] output_shapes, string name = "IteratorGetNext")
         {
             var dict = new Dictionary<string, object>();
             dict["iterator"] = iterator;
@@ -14253,7 +14253,7 @@ namespace Tensorflow.Operations
         /// <returns>
         ///    The Operation can be fetched from the resulting Tensor, by fetching the Operation property from the result.
         /// </returns>
-        public static Tensor iterator_get_next_as_optional(Tensor iterator, TF_DataType[] output_types, TensorShape[] output_shapes, string name = "IteratorGetNextAsOptional")
+        public static Tensor iterator_get_next_as_optional(Tensor iterator, TF_DataType[] output_types, Shape[] output_shapes, string name = "IteratorGetNextAsOptional")
         {
             var dict = new Dictionary<string, object>();
             dict["iterator"] = iterator;
@@ -14286,7 +14286,7 @@ namespace Tensorflow.Operations
         ///    the calling thread is not a member of the thread pool used to execute parallel
         ///    operations (e.g. in eager mode).
         /// </remarks>
-        public static Tensor[] iterator_get_next_sync(Tensor iterator, TF_DataType[] output_types, TensorShape[] output_shapes, string name = "IteratorGetNextSync")
+        public static Tensor[] iterator_get_next_sync(Tensor iterator, TF_DataType[] output_types, Shape[] output_shapes, string name = "IteratorGetNextSync")
         {
             var dict = new Dictionary<string, object>();
             dict["iterator"] = iterator;
@@ -14495,7 +14495,7 @@ namespace Tensorflow.Operations
         /// <returns>
         ///    The Operation can be fetched from the resulting Tensor, by fetching the Operation property from the result.
         /// </returns>
-        public static Tensor latency_stats_dataset(Tensor input_dataset, Tensor tag, TF_DataType[] output_types, TensorShape[] output_shapes, string name = "LatencyStatsDataset")
+        public static Tensor latency_stats_dataset(Tensor input_dataset, Tensor tag, TF_DataType[] output_types, Shape[] output_shapes, string name = "LatencyStatsDataset")
         {
             var dict = new Dictionary<string, object>();
             dict["input_dataset"] = input_dataset;
@@ -17568,7 +17568,7 @@ namespace Tensorflow.Operations
         ///    values. Each value must be a scalar. Data can be inserted into the table using
         ///    the insert operations. It does not support the initialization operation.
         /// </remarks>
-        public static Tensor mutable_dense_hash_table(Tensor empty_key, TF_DataType value_dtype, string container = null, string shared_name = null, bool? use_node_name_sharing = null, TensorShape value_shape = null, int? initial_num_buckets = null, float? max_load_factor = null, string name = "MutableDenseHashTable")
+        public static Tensor mutable_dense_hash_table(Tensor empty_key, TF_DataType value_dtype, string container = null, string shared_name = null, bool? use_node_name_sharing = null, Shape value_shape = null, int? initial_num_buckets = null, float? max_load_factor = null, string name = "MutableDenseHashTable")
         {
             var dict = new Dictionary<string, object>();
             dict["empty_key"] = empty_key;
@@ -17636,7 +17636,7 @@ namespace Tensorflow.Operations
         ///    values. Each value must be a scalar. Data can be inserted into the table using
         ///    the insert operations. It does not support the initialization operation.
         /// </remarks>
-        public static Tensor mutable_dense_hash_table_v2(Tensor empty_key, TF_DataType value_dtype, string container = null, string shared_name = null, bool? use_node_name_sharing = null, TensorShape value_shape = null, int? initial_num_buckets = null, float? max_load_factor = null, string name = "MutableDenseHashTableV2")
+        public static Tensor mutable_dense_hash_table_v2(Tensor empty_key, TF_DataType value_dtype, string container = null, string shared_name = null, bool? use_node_name_sharing = null, Shape value_shape = null, int? initial_num_buckets = null, float? max_load_factor = null, string name = "MutableDenseHashTableV2")
         {
             var dict = new Dictionary<string, object>();
             dict["empty_key"] = empty_key;
@@ -17742,7 +17742,7 @@ namespace Tensorflow.Operations
         ///    values. Each value must be a vector. Data can be inserted into the table using
         ///    the insert operations. It does not support the initialization operation.
         /// </remarks>
-        public static Tensor mutable_hash_table_of_tensors(TF_DataType key_dtype, TF_DataType value_dtype, string container = null, string shared_name = null, bool? use_node_name_sharing = null, TensorShape value_shape = null, string name = "MutableHashTableOfTensors")
+        public static Tensor mutable_hash_table_of_tensors(TF_DataType key_dtype, TF_DataType value_dtype, string container = null, string shared_name = null, bool? use_node_name_sharing = null, Shape value_shape = null, string name = "MutableHashTableOfTensors")
         {
             var dict = new Dictionary<string, object>();
             dict["key_dtype"] = key_dtype;
@@ -17794,7 +17794,7 @@ namespace Tensorflow.Operations
         ///    values. Each value must be a vector. Data can be inserted into the table using
         ///    the insert operations. It does not support the initialization operation.
         /// </remarks>
-        public static Tensor mutable_hash_table_of_tensors_v2(TF_DataType key_dtype, TF_DataType value_dtype, string container = null, string shared_name = null, bool? use_node_name_sharing = null, TensorShape value_shape = null, string name = "MutableHashTableOfTensorsV2")
+        public static Tensor mutable_hash_table_of_tensors_v2(TF_DataType key_dtype, TF_DataType value_dtype, string container = null, string shared_name = null, bool? use_node_name_sharing = null, Shape value_shape = null, string name = "MutableHashTableOfTensorsV2")
         {
             var dict = new Dictionary<string, object>();
             dict["key_dtype"] = key_dtype;
@@ -18607,7 +18607,7 @@ namespace Tensorflow.Operations
         /// <remarks>
         ///    Creates a dataset by applying optimizations to <c>input_dataset</c>.
         /// </remarks>
-        public static Tensor optimize_dataset(Tensor input_dataset, Tensor optimizations, TF_DataType[] output_types, TensorShape[] output_shapes, string name = "OptimizeDataset")
+        public static Tensor optimize_dataset(Tensor input_dataset, Tensor optimizations, TF_DataType[] output_types, Shape[] output_shapes, string name = "OptimizeDataset")
         {
             var dict = new Dictionary<string, object>();
             dict["input_dataset"] = input_dataset;
@@ -18654,7 +18654,7 @@ namespace Tensorflow.Operations
         /// <returns>
         ///    The Operation can be fetched from the resulting Tensor, by fetching the Operation property from the result.
         /// </returns>
-        public static Tensor[] optional_get_value(Tensor optional, TF_DataType[] output_types, TensorShape[] output_shapes, string name = "OptionalGetValue")
+        public static Tensor[] optional_get_value(Tensor optional, TF_DataType[] output_types, Shape[] output_shapes, string name = "OptionalGetValue")
         {
             var dict = new Dictionary<string, object>();
             dict["optional"] = optional;
@@ -19037,7 +19037,7 @@ namespace Tensorflow.Operations
         /// <remarks>
         ///    block indefinitely until data is available.
         /// </remarks>
-        public static Tensor outfeed_dequeue(TF_DataType dtype, TensorShape shape, int? device_ordinal = null, string name = "OutfeedDequeue")
+        public static Tensor outfeed_dequeue(TF_DataType dtype, Shape shape, int? device_ordinal = null, string name = "OutfeedDequeue")
         {
             var dict = new Dictionary<string, object>();
             dict["dtype"] = dtype;
@@ -19075,7 +19075,7 @@ namespace Tensorflow.Operations
         ///    tuple.  This operations will block indefinitely until data is available.
         ///    Output <c>i</c> corresponds to XLA tuple element <c>i</c>.
         /// </remarks>
-        public static Tensor[] outfeed_dequeue_tuple(TF_DataType[] dtypes, TensorShape[] shapes, int? device_ordinal = null, string name = "OutfeedDequeueTuple")
+        public static Tensor[] outfeed_dequeue_tuple(TF_DataType[] dtypes, Shape[] shapes, int? device_ordinal = null, string name = "OutfeedDequeueTuple")
         {
             var dict = new Dictionary<string, object>();
             dict["dtypes"] = dtypes;
@@ -19303,7 +19303,7 @@ namespace Tensorflow.Operations
         /// <returns>
         ///    The Operation can be fetched from the resulting Tensor, by fetching the Operation property from the result.
         /// </returns>
-        public static Tensor padded_batch_dataset(Tensor input_dataset, Tensor batch_size, Tensor[] padded_shapes, Tensor[] padding_values, TensorShape[] output_shapes, string name = "PaddedBatchDataset")
+        public static Tensor padded_batch_dataset(Tensor input_dataset, Tensor batch_size, Tensor[] padded_shapes, Tensor[] padding_values, Shape[] output_shapes, string name = "PaddedBatchDataset")
         {
             var dict = new Dictionary<string, object>();
             dict["input_dataset"] = input_dataset;
@@ -19347,7 +19347,7 @@ namespace Tensorflow.Operations
         /// <returns>
         ///    The Operation can be fetched from the resulting Tensor, by fetching the Operation property from the result.
         /// </returns>
-        public static Tensor padded_batch_dataset_v2(Tensor input_dataset, Tensor batch_size, Tensor[] padded_shapes, Tensor[] padding_values, Tensor drop_remainder, TensorShape[] output_shapes, string name = "PaddedBatchDatasetV2")
+        public static Tensor padded_batch_dataset_v2(Tensor input_dataset, Tensor batch_size, Tensor[] padded_shapes, Tensor[] padding_values, Tensor drop_remainder, Shape[] output_shapes, string name = "PaddedBatchDatasetV2")
         {
             var dict = new Dictionary<string, object>();
             dict["input_dataset"] = input_dataset;
@@ -19401,7 +19401,7 @@ namespace Tensorflow.Operations
         ///    to 0 in the shape attr.  In this case DequeueMany will pad up to the maximum
         ///    size of any given element in the minibatch.  See below for details.
         /// </remarks>
-        public static Tensor padding_f_i_f_o_queue(TF_DataType[] component_types, TensorShape[] shapes = null, int? capacity = null, string container = null, string shared_name = null, string name = "PaddingFIFOQueue")
+        public static Tensor padding_f_i_f_o_queue(TF_DataType[] component_types, Shape[] shapes = null, int? capacity = null, string container = null, string shared_name = null, string name = "PaddingFIFOQueue")
         {
             var dict = new Dictionary<string, object>();
             dict["component_types"] = component_types;
@@ -19458,7 +19458,7 @@ namespace Tensorflow.Operations
         ///    to 0 in the shape attr.  In this case DequeueMany will pad up to the maximum
         ///    size of any given element in the minibatch.  See below for details.
         /// </remarks>
-        public static Tensor padding_f_i_f_o_queue_v2(TF_DataType[] component_types, TensorShape[] shapes = null, int? capacity = null, string container = null, string shared_name = null, string name = "PaddingFIFOQueueV2")
+        public static Tensor padding_f_i_f_o_queue_v2(TF_DataType[] component_types, Shape[] shapes = null, int? capacity = null, string container = null, string shared_name = null, string name = "PaddingFIFOQueueV2")
         {
             var dict = new Dictionary<string, object>();
             dict["component_types"] = component_types;
@@ -19511,7 +19511,7 @@ namespace Tensorflow.Operations
         ///    will copy pieces of the input into the output as they become available, in
         ///    some situations this can provide a performance benefit.
         /// </remarks>
-        public static Tensor parallel_concat(Tensor[] values, TensorShape shape, string name = "ParallelConcat")
+        public static Tensor parallel_concat(Tensor[] values, Shape shape, string name = "ParallelConcat")
         {
             var dict = new Dictionary<string, object>();
             dict["values"] = values;
@@ -19729,7 +19729,7 @@ namespace Tensorflow.Operations
         ///    dense_values :
         ///    The Operation can be fetched from any of the Tensorreturned in the tuple values, by fetching the Operation property.
         /// </returns>
-        public static (Tensor[] sparse_indices, Tensor[] sparse_values, Tensor[] sparse_shapes, Tensor[] dense_values) parse_example(Tensor serialized, Tensor names, Tensor[] sparse_keys, Tensor[] dense_keys, Tensor[] dense_defaults, TF_DataType[] sparse_types, TensorShape[] dense_shapes, string name = "ParseExample")
+        public static (Tensor[] sparse_indices, Tensor[] sparse_values, Tensor[] sparse_shapes, Tensor[] dense_values) parse_example(Tensor serialized, Tensor names, Tensor[] sparse_keys, Tensor[] dense_keys, Tensor[] dense_defaults, TF_DataType[] sparse_types, Shape[] dense_shapes, string name = "ParseExample")
         {
             var dict = new Dictionary<string, object>();
             dict["serialized"] = serialized;
@@ -19801,7 +19801,7 @@ namespace Tensorflow.Operations
         /// <returns>
         ///    The Operation can be fetched from the resulting Tensor, by fetching the Operation property from the result.
         /// </returns>
-        public static Tensor parse_example_dataset(Tensor input_dataset, Tensor num_parallel_calls, Tensor[] dense_defaults, string[] sparse_keys, string[] dense_keys, TF_DataType[] sparse_types, TensorShape[] dense_shapes, TF_DataType[] output_types, TensorShape[] output_shapes, string name = "ParseExampleDataset")
+        public static Tensor parse_example_dataset(Tensor input_dataset, Tensor num_parallel_calls, Tensor[] dense_defaults, string[] sparse_keys, string[] dense_keys, TF_DataType[] sparse_types, Shape[] dense_shapes, TF_DataType[] output_types, Shape[] output_shapes, string name = "ParseExampleDataset")
         {
             var dict = new Dictionary<string, object>();
             dict["input_dataset"] = input_dataset;
@@ -19923,7 +19923,7 @@ namespace Tensorflow.Operations
         ///    feature_list_dense_lengths :
         ///    The Operation can be fetched from any of the Tensorreturned in the tuple values, by fetching the Operation property.
         /// </returns>
-        public static (Tensor[] context_sparse_indices, Tensor[] context_sparse_values, Tensor[] context_sparse_shapes, Tensor[] context_dense_values, Tensor[] feature_list_sparse_indices, Tensor[] feature_list_sparse_values, Tensor[] feature_list_sparse_shapes, Tensor[] feature_list_dense_values, Tensor[] feature_list_dense_lengths) parse_sequence_example(Tensor serialized, Tensor debug_name, Tensor[] context_dense_defaults, string[] feature_list_dense_missing_assumed_empty, string[] context_sparse_keys, string[] context_dense_keys, string[] feature_list_sparse_keys, string[] feature_list_dense_keys, int? Ncontext_sparse = null, int? Ncontext_dense = null, int? Nfeature_list_sparse = null, int? Nfeature_list_dense = null, TF_DataType[] context_sparse_types = null, TF_DataType[] feature_list_dense_types = null, TensorShape[] context_dense_shapes = null, TF_DataType[] feature_list_sparse_types = null, TensorShape[] feature_list_dense_shapes = null, string name = "ParseSequenceExample")
+        public static (Tensor[] context_sparse_indices, Tensor[] context_sparse_values, Tensor[] context_sparse_shapes, Tensor[] context_dense_values, Tensor[] feature_list_sparse_indices, Tensor[] feature_list_sparse_values, Tensor[] feature_list_sparse_shapes, Tensor[] feature_list_dense_values, Tensor[] feature_list_dense_lengths) parse_sequence_example(Tensor serialized, Tensor debug_name, Tensor[] context_dense_defaults, string[] feature_list_dense_missing_assumed_empty, string[] context_sparse_keys, string[] context_dense_keys, string[] feature_list_sparse_keys, string[] feature_list_dense_keys, int? Ncontext_sparse = null, int? Ncontext_dense = null, int? Nfeature_list_sparse = null, int? Nfeature_list_dense = null, TF_DataType[] context_sparse_types = null, TF_DataType[] feature_list_dense_types = null, Shape[] context_dense_shapes = null, TF_DataType[] feature_list_sparse_types = null, Shape[] feature_list_dense_shapes = null, string name = "ParseSequenceExample")
         {
             var dict = new Dictionary<string, object>();
             dict["serialized"] = serialized;
@@ -20029,7 +20029,7 @@ namespace Tensorflow.Operations
         ///    dense_values :
         ///    The Operation can be fetched from any of the Tensorreturned in the tuple values, by fetching the Operation property.
         /// </returns>
-        public static (Tensor[] sparse_indices, Tensor[] sparse_values, Tensor[] sparse_shapes, Tensor[] dense_values) parse_single_example(Tensor serialized, Tensor[] dense_defaults, int num_sparse, string[] sparse_keys, string[] dense_keys, TF_DataType[] sparse_types, TensorShape[] dense_shapes, string name = "ParseSingleExample")
+        public static (Tensor[] sparse_indices, Tensor[] sparse_values, Tensor[] sparse_shapes, Tensor[] dense_values) parse_single_example(Tensor serialized, Tensor[] dense_defaults, int num_sparse, string[] sparse_keys, string[] dense_keys, TF_DataType[] sparse_types, Shape[] dense_shapes, string name = "ParseSingleExample")
         {
             var dict = new Dictionary<string, object>();
             dict["serialized"] = serialized;
@@ -20140,7 +20140,7 @@ namespace Tensorflow.Operations
         ///    feature_list_dense_values :
         ///    The Operation can be fetched from any of the Tensorreturned in the tuple values, by fetching the Operation property.
         /// </returns>
-        public static (Tensor[] context_sparse_indices, Tensor[] context_sparse_values, Tensor[] context_sparse_shapes, Tensor[] context_dense_values, Tensor[] feature_list_sparse_indices, Tensor[] feature_list_sparse_values, Tensor[] feature_list_sparse_shapes, Tensor[] feature_list_dense_values) parse_single_sequence_example(Tensor serialized, Tensor feature_list_dense_missing_assumed_empty, Tensor[] context_sparse_keys, Tensor[] context_dense_keys, Tensor[] feature_list_sparse_keys, Tensor[] feature_list_dense_keys, Tensor[] context_dense_defaults, Tensor debug_name, TF_DataType[] context_sparse_types = null, TF_DataType[] feature_list_dense_types = null, TensorShape[] context_dense_shapes = null, TF_DataType[] feature_list_sparse_types = null, TensorShape[] feature_list_dense_shapes = null, string name = "ParseSingleSequenceExample")
+        public static (Tensor[] context_sparse_indices, Tensor[] context_sparse_values, Tensor[] context_sparse_shapes, Tensor[] context_dense_values, Tensor[] feature_list_sparse_indices, Tensor[] feature_list_sparse_values, Tensor[] feature_list_sparse_shapes, Tensor[] feature_list_dense_values) parse_single_sequence_example(Tensor serialized, Tensor feature_list_dense_missing_assumed_empty, Tensor[] context_sparse_keys, Tensor[] context_dense_keys, Tensor[] feature_list_sparse_keys, Tensor[] feature_list_dense_keys, Tensor[] context_dense_defaults, Tensor debug_name, TF_DataType[] context_sparse_types = null, TF_DataType[] feature_list_dense_types = null, Shape[] context_dense_shapes = null, TF_DataType[] feature_list_sparse_types = null, Shape[] feature_list_dense_shapes = null, string name = "ParseSingleSequenceExample")
         {
             var dict = new Dictionary<string, object>();
             dict["serialized"] = serialized;
@@ -20224,7 +20224,7 @@ namespace Tensorflow.Operations
         ///    intended as a way to represent a value that will always be fed, and to
         ///    provide attrs that enable the fed value to be checked at runtime.
         /// </remarks>
-        public static Tensor placeholder(TF_DataType dtype, TensorShape shape = null, string name = "Placeholder")
+        public static Tensor placeholder(TF_DataType dtype, Shape shape = null, string name = "Placeholder")
         {
             var dict = new Dictionary<string, object>();
             dict["dtype"] = dtype;
@@ -20258,7 +20258,7 @@ namespace Tensorflow.Operations
         ///    intended as a way to represent a value that will always be fed, and to
         ///    provide attrs that enable the fed value to be checked at runtime.
         /// </remarks>
-        public static Tensor placeholder_v2(TF_DataType dtype, TensorShape shape, string name = "PlaceholderV2")
+        public static Tensor placeholder_v2(TF_DataType dtype, Shape shape, string name = "PlaceholderV2")
         {
             var dict = new Dictionary<string, object>();
             dict["dtype"] = dtype;
@@ -20284,7 +20284,7 @@ namespace Tensorflow.Operations
         ///    A placeholder tensor that defaults to <c>input</c> if it is not fed.
         ///    The Operation can be fetched from the resulting Tensor, by fetching the Operation property from the result.
         /// </returns>
-        public static Tensor placeholder_with_default(Tensor input, TensorShape shape, string name = "PlaceholderWithDefault")
+        public static Tensor placeholder_with_default(Tensor input, Shape shape, string name = "PlaceholderWithDefault")
         {
             var dict = new Dictionary<string, object>();
             dict["input"] = input;
@@ -20403,7 +20403,7 @@ namespace Tensorflow.Operations
         /// <returns>
         ///    The Operation can be fetched from the resulting Tensor, by fetching the Operation property from the result.
         /// </returns>
-        public static Tensor prefetch_dataset(Tensor input_dataset, Tensor buffer_size, TF_DataType[] output_types, TensorShape[] output_shapes, string name = "PrefetchDataset")
+        public static Tensor prefetch_dataset(Tensor input_dataset, Tensor buffer_size, TF_DataType[] output_types, Shape[] output_shapes, string name = "PrefetchDataset")
         {
             var dict = new Dictionary<string, object>();
             dict["input_dataset"] = input_dataset;
@@ -20532,7 +20532,7 @@ namespace Tensorflow.Operations
         ///    and DequeueMany) on a PriorityQueue will all require (resp. output) one extra
         ///    entry in their input (resp. output) lists.
         /// </remarks>
-        public static Tensor priority_queue(TensorShape[] shapes, TF_DataType[] component_types = null, int? capacity = null, string container = null, string shared_name = null, string name = "PriorityQueue")
+        public static Tensor priority_queue(Shape[] shapes, TF_DataType[] component_types = null, int? capacity = null, string container = null, string shared_name = null, string name = "PriorityQueue")
         {
             var dict = new Dictionary<string, object>();
             dict["shapes"] = shapes;
@@ -20587,7 +20587,7 @@ namespace Tensorflow.Operations
         ///    and DequeueMany) on a PriorityQueue will all require (resp. output) one extra
         ///    entry in their input (resp. output) lists.
         /// </remarks>
-        public static Tensor priority_queue_v2(TensorShape[] shapes, TF_DataType[] component_types = null, int? capacity = null, string container = null, string shared_name = null, string name = "PriorityQueueV2")
+        public static Tensor priority_queue_v2(Shape[] shapes, TF_DataType[] component_types = null, int? capacity = null, string container = null, string shared_name = null, string name = "PriorityQueueV2")
         {
             var dict = new Dictionary<string, object>();
             dict["shapes"] = shapes;
@@ -22781,7 +22781,7 @@ namespace Tensorflow.Operations
         /// <returns>
         ///    The Operation can be fetched from the resulting Tensor, by fetching the Operation property from the result.
         /// </returns>
-        public static Tensor random_dataset(Tensor seed, Tensor seed2, TF_DataType[] output_types, TensorShape[] output_shapes, string name = "RandomDataset")
+        public static Tensor random_dataset(Tensor seed, Tensor seed2, TF_DataType[] output_types, Shape[] output_shapes, string name = "RandomDataset")
         {
             var dict = new Dictionary<string, object>();
             dict["seed"] = seed;
@@ -23035,7 +23035,7 @@ namespace Tensorflow.Operations
         ///    The handle to the queue.
         ///    The Operation can be fetched from the resulting Tensor, by fetching the Operation property from the result.
         /// </returns>
-        public static Tensor random_shuffle_queue(TF_DataType[] component_types, TensorShape[] shapes = null, int? capacity = null, int? min_after_dequeue = null, int? seed = null, int? seed2 = null, string container = null, string shared_name = null, string name = "RandomShuffleQueue")
+        public static Tensor random_shuffle_queue(TF_DataType[] component_types, Shape[] shapes = null, int? capacity = null, int? min_after_dequeue = null, int? seed = null, int? seed2 = null, string container = null, string shared_name = null, string name = "RandomShuffleQueue")
         {
             var dict = new Dictionary<string, object>();
             dict["component_types"] = component_types;
@@ -23101,7 +23101,7 @@ namespace Tensorflow.Operations
         ///    The handle to the queue.
         ///    The Operation can be fetched from the resulting Tensor, by fetching the Operation property from the result.
         /// </returns>
-        public static Tensor random_shuffle_queue_v2(TF_DataType[] component_types, TensorShape[] shapes = null, int? capacity = null, int? min_after_dequeue = null, int? seed = null, int? seed2 = null, string container = null, string shared_name = null, string name = "RandomShuffleQueueV2")
+        public static Tensor random_shuffle_queue_v2(TF_DataType[] component_types, Shape[] shapes = null, int? capacity = null, int? min_after_dequeue = null, int? seed = null, int? seed2 = null, string container = null, string shared_name = null, string name = "RandomShuffleQueueV2")
         {
             var dict = new Dictionary<string, object>();
             dict["component_types"] = component_types;
@@ -23322,7 +23322,7 @@ namespace Tensorflow.Operations
         /// <returns>
         ///    The Operation can be fetched from the resulting Tensor, by fetching the Operation property from the result.
         /// </returns>
-        public static Tensor range_dataset(Tensor start, Tensor stop, Tensor step, TF_DataType[] output_types, TensorShape[] output_shapes, string name = "RangeDataset")
+        public static Tensor range_dataset(Tensor start, Tensor stop, Tensor step, TF_DataType[] output_types, Shape[] output_shapes, string name = "RangeDataset")
         {
             var dict = new Dictionary<string, object>();
             dict["start"] = start;
@@ -24458,7 +24458,7 @@ namespace Tensorflow.Operations
         /// <returns>
         ///    The Operation can be fetched from the resulting Tensor, by fetching the Operation property from the result.
         /// </returns>
-        public static Tensor repeat_dataset(Tensor input_dataset, Tensor count, TF_DataType[] output_types, TensorShape[] output_shapes, string name = "RepeatDataset")
+        public static Tensor repeat_dataset(Tensor input_dataset, Tensor count, TF_DataType[] output_types, Shape[] output_shapes, string name = "RepeatDataset")
         {
             var dict = new Dictionary<string, object>();
             dict["input_dataset"] = input_dataset;
@@ -29811,7 +29811,7 @@ namespace Tensorflow.Operations
         /// <remarks>
         ///    pseudorandomly.
         /// </remarks>
-        public static Tensor shuffle_and_repeat_dataset(Tensor input_dataset, Tensor buffer_size, Tensor seed, Tensor seed2, Tensor count, TF_DataType[] output_types, TensorShape[] output_shapes, string name = "ShuffleAndRepeatDataset")
+        public static Tensor shuffle_and_repeat_dataset(Tensor input_dataset, Tensor buffer_size, Tensor seed, Tensor seed2, Tensor count, TF_DataType[] output_types, Shape[] output_shapes, string name = "ShuffleAndRepeatDataset")
         {
             var dict = new Dictionary<string, object>();
             dict["input_dataset"] = input_dataset;
@@ -29862,7 +29862,7 @@ namespace Tensorflow.Operations
         /// <returns>
         ///    The Operation can be fetched from the resulting Tensor, by fetching the Operation property from the result.
         /// </returns>
-        public static Tensor shuffle_dataset(Tensor input_dataset, Tensor buffer_size, Tensor seed, Tensor seed2, TF_DataType[] output_types, TensorShape[] output_shapes, bool? reshuffle_each_iteration = null, string name = "ShuffleDataset")
+        public static Tensor shuffle_dataset(Tensor input_dataset, Tensor buffer_size, Tensor seed, Tensor seed2, TF_DataType[] output_types, Shape[] output_shapes, bool? reshuffle_each_iteration = null, string name = "ShuffleDataset")
         {
             var dict = new Dictionary<string, object>();
             dict["input_dataset"] = input_dataset;
@@ -30084,7 +30084,7 @@ namespace Tensorflow.Operations
         /// <returns>
         ///    The Operation can be fetched from the resulting Tensor, by fetching the Operation property from the result.
         /// </returns>
-        public static Tensor skip_dataset(Tensor input_dataset, Tensor count, TF_DataType[] output_types, TensorShape[] output_shapes, string name = "SkipDataset")
+        public static Tensor skip_dataset(Tensor input_dataset, Tensor count, TF_DataType[] output_types, Shape[] output_shapes, string name = "SkipDataset")
         {
             var dict = new Dictionary<string, object>();
             dict["input_dataset"] = input_dataset;
@@ -30222,7 +30222,7 @@ namespace Tensorflow.Operations
         /// <returns>
         ///    The Operation can be fetched from the resulting Tensor, by fetching the Operation property from the result.
         /// </returns>
-        public static Tensor slide_dataset(Tensor input_dataset, Tensor window_size, Tensor window_shift, Tensor window_stride, TF_DataType[] output_types, TensorShape[] output_shapes, string name = "SlideDataset")
+        public static Tensor slide_dataset(Tensor input_dataset, Tensor window_size, Tensor window_shift, Tensor window_stride, TF_DataType[] output_types, Shape[] output_shapes, string name = "SlideDataset")
         {
             var dict = new Dictionary<string, object>();
             dict["input_dataset"] = input_dataset;
@@ -31745,7 +31745,7 @@ namespace Tensorflow.Operations
         ///    resets the aggregate to 0, and increments the global_step recorded by
         ///    the accumulator.
         /// </remarks>
-        public static Tensor sparse_conditional_accumulator(TF_DataType dtype, TensorShape shape, string container = null, string shared_name = null, string name = "SparseConditionalAccumulator")
+        public static Tensor sparse_conditional_accumulator(TF_DataType dtype, Shape shape, string container = null, string shared_name = null, string name = "SparseConditionalAccumulator")
         {
             var dict = new Dictionary<string, object>();
             dict["dtype"] = dtype;
@@ -33565,7 +33565,7 @@ namespace Tensorflow.Operations
         /// <returns>
         ///    The Operation can be fetched from the resulting Tensor, by fetching the Operation property from the result.
         /// </returns>
-        public static Tensor sql_dataset(Tensor driver_name, Tensor data_source_name, Tensor query, TF_DataType[] output_types, TensorShape[] output_shapes, string name = "SqlDataset")
+        public static Tensor sql_dataset(Tensor driver_name, Tensor data_source_name, Tensor query, TF_DataType[] output_types, Shape[] output_shapes, string name = "SqlDataset")
         {
             var dict = new Dictionary<string, object>();
             dict["driver_name"] = driver_name;
@@ -35743,7 +35743,7 @@ namespace Tensorflow.Operations
         /// <returns>
         ///    The Operation can be fetched from the resulting Tensor, by fetching the Operation property from the result.
         /// </returns>
-        public static Tensor take_dataset(Tensor input_dataset, Tensor count, TF_DataType[] output_types, TensorShape[] output_shapes, string name = "TakeDataset")
+        public static Tensor take_dataset(Tensor input_dataset, Tensor count, TF_DataType[] output_types, Shape[] output_shapes, string name = "TakeDataset")
         {
             var dict = new Dictionary<string, object>();
             dict["input_dataset"] = input_dataset;
@@ -35953,7 +35953,7 @@ namespace Tensorflow.Operations
         ///    var = state_ops.assign_add(var, [[6.0, 7.0]])
         ///    final = state_ops._destroy_temporary_variable(var, var_name=var_name)
         /// </remarks>
-        public static Tensor temporary_variable(TensorShape shape, TF_DataType dtype, string var_name = null, string name = "TemporaryVariable")
+        public static Tensor temporary_variable(Shape shape, TF_DataType dtype, string var_name = null, string name = "TemporaryVariable")
         {
             var dict = new Dictionary<string, object>();
             dict["shape"] = shape;
@@ -36028,7 +36028,7 @@ namespace Tensorflow.Operations
         ///    lengths :
         ///    The Operation can be fetched from any of the Tensorreturned in the tuple values, by fetching the Operation property.
         /// </returns>
-        public static (Tensor value, Tensor lengths) tensor_array_concat_v2(Tensor handle, Tensor flow_in, TF_DataType dtype, TensorShape element_shape_except0 = null, string name = "TensorArrayConcatV2")
+        public static (Tensor value, Tensor lengths) tensor_array_concat_v2(Tensor handle, Tensor flow_in, TF_DataType dtype, Shape element_shape_except0 = null, string name = "TensorArrayConcatV2")
         {
             var dict = new Dictionary<string, object>();
             dict["handle"] = handle;
@@ -36089,7 +36089,7 @@ namespace Tensorflow.Operations
         ///    
         ///    All elements must have the same shape (excepting the first dimension).
         /// </remarks>
-        public static (Tensor value, Tensor lengths) tensor_array_concat_v3(Tensor handle, Tensor flow_in, TF_DataType dtype, TensorShape element_shape_except0 = null, string name = "TensorArrayConcatV3")
+        public static (Tensor value, Tensor lengths) tensor_array_concat_v3(Tensor handle, Tensor flow_in, TF_DataType dtype, Shape element_shape_except0 = null, string name = "TensorArrayConcatV3")
         {
             var dict = new Dictionary<string, object>();
             dict["handle"] = handle;
@@ -36124,7 +36124,7 @@ namespace Tensorflow.Operations
         /// <returns>
         ///    The Operation can be fetched from the resulting Tensor, by fetching the Operation property from the result.
         /// </returns>
-        public static Tensor tensor_array_gather_v2(Tensor handle, Tensor indices, Tensor flow_in, TF_DataType dtype, TensorShape element_shape = null, string name = "TensorArrayGatherV2")
+        public static Tensor tensor_array_gather_v2(Tensor handle, Tensor indices, Tensor flow_in, TF_DataType dtype, Shape element_shape = null, string name = "TensorArrayGatherV2")
         {
             var dict = new Dictionary<string, object>();
             dict["handle"] = handle;
@@ -36169,7 +36169,7 @@ namespace Tensorflow.Operations
         /// <remarks>
         ///    All elements selected by <c>indices</c> must have the same shape.
         /// </remarks>
-        public static Tensor tensor_array_gather_v3(Tensor handle, Tensor indices, Tensor flow_in, TF_DataType dtype, TensorShape element_shape = null, string name = "TensorArrayGatherV3")
+        public static Tensor tensor_array_gather_v3(Tensor handle, Tensor indices, Tensor flow_in, TF_DataType dtype, Shape element_shape = null, string name = "TensorArrayGatherV3")
         {
             var dict = new Dictionary<string, object>();
             dict["handle"] = handle;
@@ -36613,7 +36613,7 @@ namespace Tensorflow.Operations
         /// <returns>
         ///    The Operation can be fetched from the resulting Tensor, by fetching the Operation property from the result.
         /// </returns>
-        public static Tensor tensor_array_v2(Tensor size, TF_DataType dtype, TensorShape element_shape = null, bool? dynamic_size = null, bool? clear_after_read = null, string tensor_array_name = null, string name = "TensorArrayV2")
+        public static Tensor tensor_array_v2(Tensor size, TF_DataType dtype, Shape element_shape = null, bool? dynamic_size = null, bool? clear_after_read = null, string tensor_array_name = null, string name = "TensorArrayV2")
         {
             var dict = new Dictionary<string, object>();
             dict["size"] = size;
@@ -36679,7 +36679,7 @@ namespace Tensorflow.Operations
         /// <remarks>
         ///    Write data via Write and read via Read or Pack.
         /// </remarks>
-        public static (Tensor handle, Tensor flow) tensor_array_v3(Tensor size, TF_DataType dtype, TensorShape element_shape = null, bool? dynamic_size = null, bool? clear_after_read = null, bool? identical_element_shapes = null, string tensor_array_name = null, string name = "TensorArrayV3")
+        public static (Tensor handle, Tensor flow) tensor_array_v3(Tensor size, TF_DataType dtype, Shape element_shape = null, bool? dynamic_size = null, bool? clear_after_read = null, bool? identical_element_shapes = null, string tensor_array_name = null, string name = "TensorArrayV3")
         {
             var dict = new Dictionary<string, object>();
             dict["size"] = size;
@@ -36776,7 +36776,7 @@ namespace Tensorflow.Operations
         /// <returns>
         ///    The Operation can be fetched from the resulting Tensor, by fetching the Operation property from the result.
         /// </returns>
-        public static Tensor tensor_dataset(Tensor[] components, TensorShape[] output_shapes, string name = "TensorDataset")
+        public static Tensor tensor_dataset(Tensor[] components, Shape[] output_shapes, string name = "TensorDataset")
         {
             var dict = new Dictionary<string, object>();
             dict["components"] = components;
@@ -37144,7 +37144,7 @@ namespace Tensorflow.Operations
         /// <returns>
         ///    The Operation can be fetched from the resulting Tensor, by fetching the Operation property from the result.
         /// </returns>
-        public static Tensor tensor_slice_dataset(Tensor[] components, TensorShape[] output_shapes, string name = "TensorSliceDataset")
+        public static Tensor tensor_slice_dataset(Tensor[] components, Shape[] output_shapes, string name = "TensorSliceDataset")
         {
             var dict = new Dictionary<string, object>();
             dict["components"] = components;
@@ -37899,7 +37899,7 @@ namespace Tensorflow.Operations
         /// <returns>
         ///    The Operation can be fetched from the resulting Tensor, by fetching the Operation property from the result.
         /// </returns>
-        public static Tensor unbatch_dataset(Tensor input_dataset, TF_DataType[] output_types, TensorShape[] output_shapes, string name = "UnbatchDataset")
+        public static Tensor unbatch_dataset(Tensor input_dataset, TF_DataType[] output_types, Shape[] output_shapes, string name = "UnbatchDataset")
         {
             var dict = new Dictionary<string, object>();
             dict["input_dataset"] = input_dataset;
@@ -38656,7 +38656,7 @@ namespace Tensorflow.Operations
         /// <returns>
         ///    The Operation can be fetched from the resulting Tensor, by fetching the Operation property from the result.
         /// </returns>
-        public static Tensor var_handle_op(TF_DataType dtype, TensorShape shape, string container = null, string shared_name = null, string name = "VarHandleOp")
+        public static Tensor var_handle_op(TF_DataType dtype, Shape shape, string container = null, string shared_name = null, string name = "VarHandleOp")
         {
             var dict = new Dictionary<string, object>();
             dict["dtype"] = dtype;
@@ -38710,7 +38710,7 @@ namespace Tensorflow.Operations
         /// <returns>
         ///    The Operation can be fetched from the resulting Tensor, by fetching the Operation property from the result.
         /// </returns>
-        public static Tensor variable(TensorShape shape, TF_DataType dtype, string container = null, string shared_name = null, string name = "Variable")
+        public static Tensor variable(Shape shape, TF_DataType dtype, string container = null, string shared_name = null, string name = "Variable")
         {
             var dict = new Dictionary<string, object>();
             dict["shape"] = shape;
@@ -38787,7 +38787,7 @@ namespace Tensorflow.Operations
         ///    TODO(zhifengc/mrry): Adds a pointer to a more detail document
         ///    about sharing states in tensorflow.
         /// </remarks>
-        public static Tensor variable_v2(TensorShape shape, TF_DataType dtype, string container = null, string shared_name = null, string name = "VariableV2")
+        public static Tensor variable_v2(Shape shape, TF_DataType dtype, string container = null, string shared_name = null, string name = "VariableV2")
         {
             var dict = new Dictionary<string, object>();
             dict["shape"] = shape;
@@ -39064,7 +39064,7 @@ namespace Tensorflow.Operations
         /// <returns>
         ///    The Operation can be fetched from the resulting Tensor, by fetching the Operation property from the result.
         /// </returns>
-        public static Tensor zip_dataset(Tensor[] input_datasets, TF_DataType[] output_types, TensorShape[] output_shapes, string name = "ZipDataset")
+        public static Tensor zip_dataset(Tensor[] input_datasets, TF_DataType[] output_types, Shape[] output_shapes, string name = "ZipDataset")
         {
             var dict = new Dictionary<string, object>();
             dict["input_datasets"] = input_datasets;

src/TensorFlowNET.Core/Operations/gen_resource_variable_ops.cs

@@ -94,7 +94,7 @@ namespace Tensorflow
         /// <param name="shared_name"></param>
         /// <param name="name"></param>
         /// <returns></returns>
-        public static Tensor var_handle_op(TF_DataType dtype, TensorShape shape,
+        public static Tensor var_handle_op(TF_DataType dtype, Shape shape,
             string container = "", string shared_name = "", string name = null)
         {
             if (tf.Context.executing_eagerly())

src/TensorFlowNET.Core/Operations/image_ops_impl.cs

@@ -53,11 +53,11 @@ namespace Tensorflow
 
         internal static long[] _ImageDimensions(Tensor image, int rank)
         {
-            if (image.TensorShape.is_fully_defined())
-                return image.TensorShape.as_list();
+            if (image.shape.IsFullyDefined)
+                return image.shape.dims;
             else
             {
-                var static_shape = image.TensorShape.with_rank(rank).as_list();
+                var static_shape = image.shape.with_rank(rank).dims;
                 var dynamic_shape = array_ops.unstack(array_ops.shape(image), rank);
 
                 long[] ss_storage = null;
@@ -82,23 +82,23 @@ namespace Tensorflow
 
         internal static Operation[] _CheckAtLeast3DImage(Tensor image, bool require_static)
         {
-            TensorShape image_shape;
+            Shape image_shape;
             try
             {
-                if (image.TensorShape.ndim == Unknown)
+                if (image.shape.ndim == Unknown)
                 {
-                    image_shape = image.TensorShape.with_rank(3);
+                    image_shape = image.shape.with_rank(3);
                 }
                 else
                 {
-                    image_shape = image.TensorShape.with_rank_at_least(3);
+                    image_shape = image.shape.with_rank_at_least(3);
                 }
             }
             catch (ValueError)
             {
                 throw new ValueError("'image' must be at least three-dimensional.");
             }
-            if (require_static & !image_shape.is_fully_defined())
+            if (require_static & !image_shape.IsFullyDefined)
             {
                 throw new ValueError("\'image\' must be fully defined.");
             }
@@ -110,11 +110,11 @@ namespace Tensorflow
                 }
             }
 
-            var image_shape_last_three_elements = new TensorShape(new[] {
+            var image_shape_last_three_elements = new Shape(new[] {
                                                 image_shape.dims[image_shape.dims.Length - 1],
                                                 image_shape.dims[image_shape.dims.Length - 2],
                                                 image_shape.dims[image_shape.dims.Length - 3]});
-            if (!image_shape_last_three_elements.is_fully_defined())
+            if (!image_shape_last_three_elements.IsFullyDefined)
             {
                 Tensor image_shape_ = array_ops.shape(image);
                 var image_shape_return = tf.constant(new[] {
@@ -142,15 +142,15 @@ namespace Tensorflow
 
         internal static Tensor fix_image_flip_shape(Tensor image, Tensor result)
         {
-            TensorShape image_shape = image.shape;
+            Shape image_shape = image.shape;
             if (image_shape == image_shape.unknown_shape())
             {
                 // c# defaults null types to 0 anyhow, so this should be a pretty equivalent port
-                result.set_shape(new TensorShape(new long[] { 0, 0, 0 }));
+                result.shape = new long[] { 0, 0, 0 };
             }
             else
             {
-                result.set_shape(image_shape);
+                result.shape = image_shape;
             }
             return result;
         }
@@ -173,7 +173,7 @@ namespace Tensorflow
              {
                  image = ops.convert_to_tensor(image, name: "image");
                  image = _AssertAtLeast3DImage(image);
-                 TensorShape shape = image.shape;
+                 Shape shape = image.shape;
                  if (shape.ndim == 3 || shape.ndim == Unknown)
                  {
                      Tensor uniform_random = random_ops.random_uniform(new int[] { }, 0f, 1.0f, seed: seed);
@@ -219,7 +219,7 @@ namespace Tensorflow
               {
                   image = ops.convert_to_tensor(image, name: "image");
                   image = _AssertAtLeast3DImage(image);
-                  TensorShape shape = image.shape;
+                  Shape shape = image.shape;
                   if (shape.ndim == 3 || shape.ndim == Unknown)
                   {
                       return fix_image_flip_shape(image, gen_array_ops.reverse(image, new { flip_index }));
@@ -245,10 +245,10 @@ namespace Tensorflow
                  // can't get k to convert to tensor without throwing error about it being an int---
                  // might rework later. for now, k2 == k as Tensor
                  Tensor k2 = ops.convert_to_tensor(k, dtype: dtypes.int32, name: "k");
-                 k2.TensorShape.assert_has_rank(0);
+                 k2.shape.assert_has_rank(0);
                  k2 = gen_ops.mod(k2, tf.constant(4));
 
-                 TensorShape shape = image.shape;
+                 Shape shape = image.shape;
                  if (shape.ndim == 3 || shape.ndim == Unknown)
                  {
                      return _rot90_3D(image, k, scope);
@@ -284,7 +284,7 @@ namespace Tensorflow
                                 math_ops.equal(k, 3), _rot270()};
 
             var result = control_flow_ops.case_v2(cases, callable_default: () => new Tensor[] { image }, exclusive: true, name: name_scope);
-            result.set_shape(new long[] { -1, -1, image.TensorShape.dims[2] });
+            result.shape = new long[] { -1, -1, image.shape.dims[2] };
             return result;
         }
 
@@ -295,7 +295,7 @@ namespace Tensorflow
                  {
                      image = ops.convert_to_tensor(image, name: "image");
                      image = _AssertAtLeast3DImage(image);
-                     TensorShape shape = image.shape;
+                     Shape shape = image.shape;
                      if (shape.ndim == 3 || shape.ndim == Unknown)
                      {
                          return array_ops.transpose(image, new[] { 1, 0, 2 }, name: name);
@@ -322,14 +322,14 @@ namespace Tensorflow
                     return image;
 
                 _AssertAtLeast3DImage(image);
-                var rank = image.TensorShape.ndim;
+                var rank = image.shape.ndim;
                 if (rank != 3 && rank != 4)
                     throw new ValueError(String.Format(@"`image` should either be a Tensor with rank = 3
 or rank = 4. Had rank = {0}", rank));
 
                 object[] _get_dim(Tensor tensor, int idx)
                 {
-                    var static_shape = tensor.TensorShape.dims[idx];
+                    var static_shape = tensor.shape.dims[idx];
                     if (static_shape != (int)None)
                         return new object[2] { static_shape, false };
                     return new object[2] { array_ops.shape(tensor)[idx], true };
@@ -445,7 +445,7 @@ or rank = 4. Had rank = {0}", rank));
                  image = ops.convert_to_tensor(image, name: "image");
 
                  bool is_batch = true;
-                 TensorShape image_shape = image.shape;
+                 Shape image_shape = image.shape;
                  if (image_shape.ndim == 3)
                  {
                      is_batch = false;
@@ -455,7 +455,7 @@ or rank = 4. Had rank = {0}", rank));
                  {
                      is_batch = false;
                      image = array_ops.expand_dims(image, 0);
-                     image.set_shape(new TensorShape(0, 0, 0, 0));
+                     image.shape = new Shape(0, 0, 0, 0);
                  }
                  else if (image_shape.ndim != 4)
                  {
@@ -494,7 +494,7 @@ or rank = 4. Had rank = {0}", rank));
                  );
                  var padded = array_ops.pad(image, paddings);
 
-                 TensorShape padded_shape_result()
+                 Shape padded_shape_result()
                  {
                      long[] i_remnants = { };
                      foreach (var i in new[] { bhwd[0], target_height, target_width, bhwd[3] })
@@ -502,10 +502,10 @@ or rank = 4. Had rank = {0}", rank));
                              return null;
                          else
                              i_remnants[i_remnants.Length] = i;
-                     return new TensorShape(i_remnants);
+                     return new Shape(i_remnants);
                  };
-                 TensorShape padded_shape = padded_shape_result();
-                 padded.set_shape(padded_shape);
+                 Shape padded_shape = padded_shape_result();
+                 padded.shape = padded_shape;
 
                  if (!is_batch)
                  {
@@ -524,7 +524,7 @@ or rank = 4. Had rank = {0}", rank));
                  image = ops.convert_to_tensor(image, name: "image");
 
                  bool is_batch = true;
-                 TensorShape image_shape = image.shape;
+                 Shape image_shape = image.shape;
                  if (image_shape.ndim == 3)
                  {
                      is_batch = false;
@@ -534,7 +534,7 @@ or rank = 4. Had rank = {0}", rank));
                  {
                      is_batch = false;
                      image = array_ops.expand_dims(image, 0);
-                     image.set_shape(new TensorShape(new long[] { 0, 0, 0, 0 }));
+                     image.shape = new long[] { 0, 0, 0, 0 };
                  }
                  else if (image_shape.ndim != 4)
                  {
@@ -573,7 +573,7 @@ or rank = 4. Had rank = {0}", rank));
                      image, array_ops.stack(new[] { 0, offset_height, offset_width, 0 }),
                      array_ops.stack(new[] { -1, target_height, target_width, -1 }));
 
-                 TensorShape cropped_shape_result()
+                 Shape cropped_shape_result()
                  {
                      long[] i_remnants = { };
                      foreach (var i in new[] { bhwd[0], target_height, target_width, bhwd[3] })
@@ -581,10 +581,10 @@ or rank = 4. Had rank = {0}", rank));
                              return null;
                          else
                              i_remnants[i_remnants.Length] = i;
-                     return new TensorShape(i_remnants);
+                     return new Shape(i_remnants);
                  };
                  var cropped_shape = cropped_shape_result();
-                 cropped.set_shape(cropped_shape);
+                 cropped.shape = cropped_shape;
 
                  if (!is_batch)
                  {
@@ -601,7 +601,7 @@ or rank = 4. Had rank = {0}", rank));
                 return tf_with(ops.name_scope(null, "resize_image_with_crop_or_pad", new[] { image }), delegate
                  {
                      image = ops.convert_to_tensor(image, name: "image");
-                     TensorShape image_shape = image.shape;
+                     Shape image_shape = image.shape;
                      bool is_batch = true;
                      if (image_shape.ndim == 3)
                      {
@@ -612,7 +612,7 @@ or rank = 4. Had rank = {0}", rank));
                      {
                          is_batch = false;
                          image = array_ops.expand_dims(image, 0);
-                         image.set_shape(new TensorShape(new int[] { 0, 0, 0, 0 }));
+                         image.shape = new long[] { 0, 0, 0, 0 };
                      }
                      else if (image_shape.ndim != 4)
                      {
@@ -684,7 +684,7 @@ or rank = 4. Had rank = {0}", rank));
                      Tensor resized = pad_to_bounding_box(cropped, offset_pad_height, offset_pad_width,
                                                      (int)target_height, (int)target_width);
 
-                     if (resized.TensorShape.ndim == Unknown)
+                     if (resized.shape.ndim == Unknown)
                          throw new ValueError("resized contains no shape.");
 
                      var _rhrw_ = _ImageDimensions(resized, rank: 4);
@@ -713,20 +713,20 @@ or rank = 4. Had rank = {0}", rank));
         {
             return tf_with(ops.name_scope(name, "resize", new[] { images, size }), delegate
               {
-                  if (images.TensorShape.ndim == Unknown)
+                  if (images.shape.ndim == Unknown)
                       throw new ValueError("\'images\' contains no shape.");
                   bool is_batch = true;
-                  if (images.TensorShape.ndim == 3)
+                  if (images.shape.ndim == 3)
                   {
                       is_batch = false;
                       images = array_ops.expand_dims(images, 0);
                   }
-                  else if (images.TensorShape.ndim != 4)
+                  else if (images.shape.ndim != 4)
                       throw new ValueError("\'images\' must have either 3 or 4 dimensions.");
 
                   var (height, width) = (images.dims[1], images.dims[2]);
 
-                  if (!size.TensorShape.is_compatible_with(new[] { 2 }))
+                  if (!size.shape.is_compatible_with(new[] { 2 }))
                       throw new ValueError(@"\'size\' must be a 1-D Tensor of 2 elements:
 new_height, new_width");
 
@@ -782,7 +782,7 @@ new_height, new_width");
 
                   images = resizer_fn(images, size);
 
-                  images.set_shape(new TensorShape(new long[] { Unknown, new_height_const, new_width_const, Unknown }));
+                  images.shape = new Shape(Unknown, new_height_const, new_width_const, Unknown);
 
                   if (!is_batch)
                       images = array_ops.squeeze(images, axis: new int[] { 0 });
@@ -862,7 +862,7 @@ new_height, new_width");
                 return tf_with(ops.name_scope(null, "resize_image_with_pad", new[] { image }), delegate
                  {
                      image = ops.convert_to_tensor(image, name: "tensor");
-                     var image_shape = image.TensorShape;
+                     var image_shape = image.shape;
                      bool is_batch = true;
                      if (image_shape.ndim == 3)
                      {
@@ -873,7 +873,7 @@ new_height, new_width");
                      {
                          is_batch = false;
                          image = array_ops.expand_dims(image, 0);
-                         image.set_shape(new TensorShape(new[] { Unknown, Unknown, Unknown, Unknown }));
+                         image.shape = new Shape(Unknown, Unknown, Unknown, Unknown);
                      }
                      else if (image_shape.ndim != 4)
                      {
@@ -928,7 +928,7 @@ new_height, new_width");
                      var padded = pad_to_bounding_box(resized, p_height, p_width, target_height,
                                                      target_width);
 
-                     if (padded.TensorShape.ndim == Unknown)
+                     if (padded.shape.ndim == Unknown)
                          throw new ValueError("padded contains no shape.");
 
                      _ImageDimensions(padded, rank: 4);
@@ -1175,7 +1175,7 @@ new_height, new_width");
             return tf_with(ops.name_scope(name, "adjust_jpeg_quality", new[] { image }), delegate
              {
                  image = ops.convert_to_tensor(image, name: "image");
-                 var channels = image.TensorShape.as_list()[image.TensorShape.dims.Length - 1];
+                 var channels = image.shape[image.shape.dims.Length - 1];
                  var orig_dtype = image.dtype;
                  // python code checks to ensure jpeq_quality is a tensor; unnecessary here since
                  // it is passed as a tensor
@@ -1327,7 +1327,7 @@ new_height, new_width");
                                                     {0.587f, -0.27455667f, -0.52273617f},
                                                     {0.114f, -0.32134392f, 0.31119955f}};
             Tensor kernel = ops.convert_to_tensor(_rgb_to_yiq_kernel, dtype: images.dtype, name: "kernel");
-            var ndims = images.TensorShape.ndim;
+            var ndims = images.shape.ndim;
             return math_ops.tensordot(images, kernel, axes: new int[] { ndims - 1, 0 });
         }
 
@@ -1338,7 +1338,7 @@ new_height, new_width");
                                                     {0.95598634f, -0.27201283f, -1.10674021f},
                                                     {0.6208248f, -0.64720424f, 1.70423049f}};
             Tensor kernel = ops.convert_to_tensor(_yiq_to_rgb_kernel, dtype: images.dtype, name: "kernel");
-            var ndims = images.TensorShape.ndim;
+            var ndims = images.shape.ndim;
             return math_ops.tensordot(images, kernel, axes: new int[] { ndims - 1, 0 });
         }
 
@@ -1349,7 +1349,7 @@ new_height, new_width");
                                                     {0.587f, -0.28886916f, -0.51496512f},
                                                     {0.114f, 0.43601035f, -0.10001026f}};
             Tensor kernel = ops.convert_to_tensor(_rgb_to_yuv_kernel, dtype: images.dtype, name: "kernel");
-            var ndims = images.TensorShape.ndim;
+            var ndims = images.shape.ndim;
             return math_ops.tensordot(images, kernel, axes: new int[] { ndims - 1, 0 });
         }
 
@@ -1360,15 +1360,15 @@ new_height, new_width");
                                                     {0f, -0.394642334f, 2.03206185f},
                                                     {1.13988303f, -0.58062185f, 0f}};
             Tensor kernel = ops.convert_to_tensor(_yuv_to_rgb_kernel, dtype: images.dtype, name: "kernel");
-            var ndims = images.TensorShape.ndim;
+            var ndims = images.shape.ndim;
             return math_ops.tensordot(images, kernel, axes: new int[] { ndims - 1, 0 });
         }
 
         internal static (Tensor, Tensor, Operation[]) _verify_compatible_image_shapes(Tensor img1, Tensor img2)
         {
-            TensorShape shape1 = img1.TensorShape.with_rank_at_least(3);
-            TensorShape shape2 = img2.TensorShape.with_rank_at_least(3);
-            shape1 = new TensorShape(shape1.dims.Skip(shape1.dims.Length - 3).Take(shape1.dims.Length - (shape1.dims.Length - 3)).ToArray());
+            Shape shape1 = img1.shape.with_rank_at_least(3);
+            Shape shape2 = img2.shape.with_rank_at_least(3);
+            shape1 = new Shape(shape1.dims.Skip(shape1.dims.Length - 3).Take(shape1.dims.Length - (shape1.dims.Length - 3)).ToArray());
             tensor_shape.assert_is_compatible_with(self: new Tensor(shape1.dims), other: new Tensor(shape2.dims.Skip(shape2.dims.Length - 3).Take(shape2.dims.Length - (shape2.dims.Length - 3)).ToArray()));
 
             if (shape1.ndim != -1 && shape2.ndim != -1)
@@ -1653,7 +1653,7 @@ new_height, new_width");
 
         public static (Tensor, Tensor) image_gradients(Tensor image)
         {
-            if (image.TensorShape.ndim != 4)
+            if (image.shape.ndim != 4)
                 throw new ValueError(String.Format(@"image_gradients expects a 4D tensor [batch_size, h, w, d], not {0}.", image.shape));
 
             var image_shape = array_ops.shape(image);
@@ -1674,7 +1674,7 @@ new_height, new_width");
 
         public static Tensor sobel_edges(Tensor image)
         {
-            var static_image_shape = image.TensorShape;
+            var static_image_shape = image.shape;
             var image_shape = array_ops.shape(image);
             var kernels = new Tensor(new int[,] {{-1, -2, -1}, {0, 0, 0}, {1, 2, 1},
                                                  {-1, 0, 1}, {-2, 0, 2}, {-1, 0, 1}});
@@ -1695,7 +1695,7 @@ new_height, new_width");
 
             var shape = array_ops.concat(new Tensor[] { image_shape, ops.convert_to_tensor(num_kernels) }, 0);
             output = array_ops.reshape(output, shape: shape);
-            output.set_shape(static_image_shape.concatenate(new int[] { num_kernels }));
+            output.shape = static_image_shape.concatenate(new int[] { num_kernels });
             return output;
         }
 
@@ -1930,7 +1930,7 @@ new_height, new_width");
                 return tf_with(ops.name_scope(name, "non_max_suppression_padded"), delegate
                 {
                     if (!pad_to_max_output_size)
-                        if (boxes.TensorShape.rank != -1 && boxes.TensorShape.rank > 2)
+                        if (boxes.shape.ndim != -1 && boxes.shape.ndim > 2)
                             throw new ValueError(String.Format(
                                 "'pad_to_max_output_size' (value {0}) must be true for 'batched input'", pad_to_max_output_size));
                     if (name == null)
@@ -1943,11 +1943,11 @@ new_height, new_width");
                         //   0, slice(None, num_valid, None)
                         // which is what I tried to replicate below, but i don't think that Unknown is the exact
                         // equivalent to None, and don't know about the slice function bit.
-                        idx = idx[0, slice(Unknown, num_valid.TensorShape.ndim, Unknown).ToArray()[0]];
+                        idx = idx[0, slice(Unknown, num_valid.shape.ndim, Unknown).ToArray()[0]];
                     else
                     {
                         var batch_dims = array_ops.concat(new Tensor[] {
-                            new Tensor(array_ops.shape(boxes).dims.Take(boxes.TensorShape.dims.Length - 2).ToArray()),
+                            new Tensor(array_ops.shape(boxes).dims.Take(boxes.shape.dims.Length - 2).ToArray()),
                             array_ops.expand_dims(max_output_size, 0)
                         }, 0);
                         idx = array_ops.reshape(idx, batch_dims);
@@ -1984,8 +1984,8 @@ new_height, new_width");
                 return (sorted_scores, sorted_boxes, sorted_scores_indices);
             }
 
-            var batch_dims = array_ops.shape(boxes).dims.Take(boxes.TensorShape.dims.Length - 2).ToArray();
-            var num_boxes = array_ops.shape(boxes).dims[boxes.TensorShape.dims.Length - 2];
+            var batch_dims = array_ops.shape(boxes).dims.Take(boxes.shape.dims.Length - 2).ToArray();
+            var num_boxes = array_ops.shape(boxes).dims[boxes.shape.dims.Length - 2];
             boxes = array_ops.reshape(boxes, new[] { -1, num_boxes, 4 });
             scores = array_ops.reshape(scores, new[] { -1, num_boxes });
             var batch_size = array_ops.shape(boxes).dims[0];
@@ -2059,14 +2059,14 @@ new_height, new_width");
                 body: (Tensor[] args) => suppression_loop_body(args),
                 loop_vars: new object[] {
                     boxes, iou_threshold,
-                    array_ops.zeros(new TensorShape(batch_size), dtypes.int32),
+                    array_ops.zeros(new Shape(batch_size), dtypes.int32),
                     constant_op.constant(0)
                 },
-                shape_invariants: new TensorShape[] {
-                    new TensorShape(new int[] {Unknown, Unknown, 4}),
-                    new TensorShape(new int[] {}),
-                    new TensorShape(new int[] {Unknown}),
-                    new TensorShape(new int[] {})
+                shape_invariants: new Shape[] {
+                    new Shape(new int[] {Unknown, Unknown, 4}),
+                    new Shape(new int[] {}),
+                    new Shape(new int[] {Unknown}),
+                    new Shape(new int[] {})
                 }
             );
             */

src/TensorFlowNET.Core/Operations/linalg_ops.cs

@@ -7,7 +7,7 @@ namespace Tensorflow
     {
         public Tensor eye(int num_rows,
             int num_columns = -1,
-            TensorShape batch_shape = null,
+            Shape batch_shape = null,
             TF_DataType dtype = TF_DataType.TF_DOUBLE,
             string name = null)
         {
@@ -19,7 +19,7 @@ namespace Tensorflow
                 bool is_square = num_columns == num_rows;
                 var diag_size = Math.Min(num_rows, num_columns);
                 if (batch_shape == null)
-                    batch_shape = new TensorShape(new int[0]);
+                    batch_shape = new Shape(new int[0]);
                 var diag_shape = batch_shape.dims.concat(new long[] { diag_size });
 
                 long[] shape = null;

src/TensorFlowNET.Core/Operations/map_fn.cs

@@ -123,16 +123,16 @@ namespace Tensorflow
                     maximum_iterations: tf.constant(n));
                 var results_flat = r_a.Accs_ta.Select(r => r.stack()).ToArray();
 
-                var n_static = new Dimension(tensor_shape.dimension_value(elems_flat[0].TensorShape.with_rank_at_least(1).dims[0]));
+                var n_static = new Dimension(tensor_shape.dimension_value(elems_flat[0].shape.with_rank_at_least(1).dims[0]));
 
                 foreach (var elem in elems_flat.Skip(1))
                 {
-                    n_static.merge_with(new Dimension(tensor_shape.dimension_value(elem.TensorShape.with_rank_at_least(1).dims[0])));
+                    n_static.merge_with(new Dimension(tensor_shape.dimension_value(elem.shape.with_rank_at_least(1).dims[0])));
                 }
 
                 foreach (Tensor r in results_flat)
                 {
-                    r.set_shape(new TensorShape(n_static).concatenate(r.dims.Skip(1).ToArray()));
+                    r.shape = new Shape(n_static).concatenate(r.dims.Skip(1).ToArray());
                 }
 
                 // todo get working when the above caching_device is fixed

src/TensorFlowNET.Core/Operations/math_ops.cs

@@ -422,25 +422,33 @@ namespace Tensorflow
         public static Tensor lgamma(Tensor x, string name = null)
             => gen_math_ops.lgamma(x, name: name);
 
-        public static Tensor linspace(Tensor start, Tensor stop, Tensor num, string name = null, int axis = 0)
+        public static Tensor linspace(Tensor start, Tensor stop, int num = 50, string name = null, int axis = 0)
         {
             return tf_with(ops.name_scope(name, "linspace", new { start, stop }), scope =>
             {
+                var num_int_tensor = array_ops.constant(num);
+                var num_tensor = array_ops.constant(num, dtype: start.dtype);
+
+                var broadcast_shape = array_ops.broadcast_dynamic_shape(array_ops.shape(start), array_ops.shape(stop));
+                start = gen_array_ops.broadcast_to(start, broadcast_shape);
+                stop = gen_array_ops.broadcast_to(stop, broadcast_shape);
+
                 var expanded_start = array_ops.expand_dims(start, axis: axis);
                 var expanded_stop = array_ops.expand_dims(stop, axis: axis);
+
                 var shape = array_ops.shape(expanded_start);
                 var ndims = array_ops.shape(shape)[0];
 
                 var axis_tensor = array_ops.where_v2(constant_op.constant(axis >= 0), x: axis, y: ndims + axis);
 
                 // The purpose is to avoid having negative values when repeating.
-                var num_fill = gen_math_ops.maximum(num - 2, 0);
-                var n_steps = gen_math_ops.maximum(num - 1, 1);
+                var num_fill = gen_math_ops.maximum(num_int_tensor - 2, 0);
+                var n_steps = gen_math_ops.maximum(num_int_tensor - 1, 1);
                 var delta = (expanded_stop - expanded_start) / cast(n_steps, expanded_stop.dtype);
 
-                var range_end = array_ops.where_v2(num >= 0, n_steps, -1);
+                var range_end = array_ops.where_v2(num_int_tensor >= 0, n_steps, -1);
                 var desired_range = cast(range(1, range_end, dtype: dtypes.int64), delta.dtype);
-                var mask = gen_math_ops.equal(axis, range(ndims));
+                var mask = gen_math_ops.equal(axis_tensor, range(ndims));
                 var desired_range_shape = array_ops.where_v2(mask, num_fill, 1);
                 desired_range = array_ops.reshape(desired_range, desired_range_shape);
                 var res = expanded_start + delta * desired_range;
@@ -450,7 +458,7 @@ namespace Tensorflow
                 var all_tensors = new[] { expanded_start, res, expanded_stop };
                 var concatenated = array_ops.concat(all_tensors, axis: axis);
                 var begin = array_ops.zeros_like(shape);
-                var size = array_ops.where_v2(mask, num, shape);
+                var size = array_ops.where_v2(mask, num_int_tensor, shape);
 
                 return array_ops.slice(concatenated, begin, size);
             });
@@ -745,7 +753,7 @@ namespace Tensorflow
                 return tf.Context.ExecuteOp("Pow", name, new ExecuteOpArgs(x_tensor, y_tensor));
             });
 
-        public static Tensor range(object start, object limit = null, object delta = null, TF_DataType dtype = TF_DataType.DtInvalid, string name = "range")
+        public static Tensor range(object start, object limit = null, object delta = null, TF_DataType? dtype = null, string name = "range")
         {
             if (limit == null)
             {
@@ -753,36 +761,14 @@ namespace Tensorflow
                 start = 0;
             }
 
-            if (dtype == TF_DataType.DtInvalid)
-            {
-                if (limit is Tensor tensor)
-                    dtype = tensor.dtype;
-                else
-                    dtype = limit.GetType().as_tf_dtype();
-            }
-                
-
-            if (delta == null)
-            {
-                if (dtype == TF_DataType.TF_INT32)
-                    delta = 1;
-                else if (dtype == TF_DataType.TF_INT64)
-                    delta = 1L;
-                else if (dtype == TF_DataType.TF_FLOAT)
-                    delta = 1.0f;
-                else if (dtype == TF_DataType.TF_DOUBLE)
-                    delta = 1.0d;
-                else
-                    delta = 1;
-            }
+            var dtype1 = dtype ?? limit.GetDataType();
 
             return tf_with(ops.name_scope(name, "Range", new { start, limit, delta }), scope =>
             {
                 name = scope;
-                var start1 = ops.convert_to_tensor(start, name: "start", dtype: dtype);
-                var limit1 = ops.convert_to_tensor(limit, name: "limit", dtype: dtype);
-                var delta1 = ops.convert_to_tensor(delta, name: "delta", dtype: dtype);
-
+                var start1 = ops.convert_to_tensor(start, name: "start", dtype: dtype1);
+                var limit1 = ops.convert_to_tensor(limit, name: "limit", dtype: dtype1);
+                var delta1 = ops.convert_to_tensor(delta ?? 1, name: "delta", dtype: dtype1);
                 return gen_math_ops.range(start1, limit1, delta1, name);
             });
         }
@@ -860,7 +846,7 @@ namespace Tensorflow
              Tensor maxlength = null,
              TF_DataType dtype = TF_DataType.TF_INT32,
              string name = null,
-             TensorShape axis = null,
+             Shape axis = null,
              bool binary_output = false)
             => tf_with(ops.name_scope(name, "bincount"), scope =>
             {
@@ -906,9 +892,9 @@ namespace Tensorflow
         {
             Tensor _tensordot_reshape(Tensor a, int[] axes, bool flipped = false)
             {
-                if (a.TensorShape.is_fully_defined() && isinstance(axes, (typeof(List<object>), typeof(Tuple))))
+                if (a.shape.IsFullyDefined && isinstance(axes, (typeof(List<object>), typeof(Tuple))))
                 {
-                    var shape_a = a.TensorShape.as_list();
+                    var shape_a = a.shape.dims;
 
                     // axes
                     int iter = 0;
@@ -950,11 +936,11 @@ namespace Tensorflow
                                                                  + ops.convert_to_tensor(list(axes));
 
                     // new_shape
-                    TensorShape new_shape;
+                    Shape new_shape;
                     if (flipped)
-                        new_shape = new TensorShape(new int[] { prod_axes, prod_free });
+                        new_shape = new Shape(new int[] { prod_axes, prod_free });
                     else
-                        new_shape = new TensorShape(new int[] { prod_free, prod_axes });
+                        new_shape = new Shape(new int[] { prod_free, prod_axes });
                 }
 
                 throw new NotImplementedException("_tensordot_reshape");

src/TensorFlowNET.Core/Operations/nn_impl.py.cs

@@ -25,11 +25,11 @@ namespace Tensorflow
         public static Tensor conv2d_transpose(Tensor value = null,
             IVariableV1 filter = null,
             Tensor output_shape = null,
-            TensorShape strides = null,
+            Shape strides = null,
             string padding = "SAME",
             string data_format = "NHWC",
             string name = null,
-            TensorShape dilations = null)
+            Shape dilations = null)
         {
             if (dilations == null)
                 dilations = (1, 1, 1, 1);
@@ -186,7 +186,7 @@ namespace Tensorflow
         {
             return tf_with(ops.name_scope("count_nonzero", "count_nonzero", new { input_tensor }), scope =>
             {
-                var zero = array_ops.zeros(new Shape(), dtype: input_tensor.dtype);
+                var zero = array_ops.zeros(Shape.Null, dtype: input_tensor.dtype);
                 var nonzero_count = math_ops.reduce_sum(
                 math_ops.cast(gen_math_ops.not_equal(input_tensor, zero), dtype: dtype), name: "nonzero_count");
                 return nonzero_count;
@@ -200,7 +200,7 @@ namespace Tensorflow
                 name = scope;
                 logits = ops.convert_to_tensor(logits, name: "logits");
                 labels = ops.convert_to_tensor(labels, name: "labels");
-                labels.TensorShape.merge_with(logits.TensorShape);
+                labels.shape.merge_with(logits.shape);
 
                 var zeros = array_ops.zeros_like(logits, dtype: logits.dtype);
                 var cond = (logits >= zeros);

src/TensorFlowNET.Core/Operations/nn_ops.cs

@@ -96,7 +96,7 @@ namespace Tensorflow
                 var keep_mask = random_tensor >= rate;
                 ret = x * scale * math_ops.cast(keep_mask, x.dtype);
                 if (!tf.executing_eagerly())
-                    ret.set_shape(x.TensorShape);
+                    ret.shape = x.shape;
                 return ret;
             });
         }
@@ -203,14 +203,14 @@ namespace Tensorflow
                 var precise_logits = logits.dtype == TF_DataType.TF_HALF ? math_ops.cast(logits, dtypes.float32) : logits;
 
                 // Store label shape for result later.
-                var labels_static_shape = labels.TensorShape;
+                var labels_static_shape = labels.shape;
                 var labels_shape = array_ops.shape(labels);
                 /*bool static_shapes_fully_defined = (
                     labels_static_shape.is_fully_defined() &&
                         logits.get_shape()[:-1].is_fully_defined());*/
 
                 // Check if no reshapes are required.
-                if (logits.TensorShape.ndim == 2)
+                if (logits.shape.ndim == 2)
                 {
                     var (cost, _) = gen_nn_ops.sparse_softmax_cross_entropy_with_logits(
                         precise_logits, labels, name: name);
@@ -236,7 +236,7 @@ namespace Tensorflow
                 name = scope;
                 var precise_logits = logits;
                 var input_rank = array_ops.rank(precise_logits);
-                var shape = logits.TensorShape;
+                var shape = logits.shape;
 
                 if (axis != -1)
                     throw new NotImplementedException("softmax_cross_entropy_with_logits_v2_helper axis != -1");
@@ -282,7 +282,7 @@ namespace Tensorflow
             // Set output shape if known.
             if (!tf.Context.executing_eagerly())
             {
-                var shape = logits.TensorShape;
+                var shape = logits.shape;
                 if (shape != null && shape.ndim > 0)
                 {
                     var product = 1L;

src/TensorFlowNET.Core/Operations/random_ops.cs

@@ -30,7 +30,7 @@ namespace Tensorflow
         /// <param name="seed"></param>
         /// <param name="name"></param>
         /// <returns></returns>
-        public static Tensor random_normal(TensorShape shape,
+        public static Tensor random_normal(Shape shape,
             float mean = 0.0f,
             float stddev = 1.0f,
             TF_DataType dtype = TF_DataType.TF_FLOAT,

src/TensorFlowNET.Core/Operations/resource_variable_ops.cs

@@ -48,7 +48,7 @@ namespace Tensorflow
         /// <param name="name"></param>
         /// <param name="graph_mode"></param>
         /// <returns></returns>
-        public static Tensor eager_safe_variable_handle(Tensor initial_value, TensorShape shape,
+        public static Tensor eager_safe_variable_handle(Tensor initial_value, Shape shape,
             string shared_name, string name, bool graph_mode)
         {
             var dtype = initial_value.dtype.as_base_dtype();
@@ -66,7 +66,7 @@ namespace Tensorflow
         /// <param name="graph_mode"></param>
         /// <param name="initial_value"></param>
         /// <returns></returns>
-        public static Tensor variable_handle_from_shape_and_dtype(TensorShape shape, TF_DataType dtype,
+        public static Tensor variable_handle_from_shape_and_dtype(Shape shape, TF_DataType dtype,
             string shared_name, string name, bool graph_mode, Tensor initial_value = null)
         {
             var container = ops.get_default_graph().Container;
@@ -161,7 +161,7 @@ namespace Tensorflow
                 var data = new HandleData();
                 data.ShapeAndType.Add(new HandleShapeAndType
                 {
-                    Shape = handle.TensorShape.as_proto(),
+                    Shape = handle.shape.as_shape_proto(),
                     Dtype = handle.dtype.as_datatype_enum()
                 });
                 return data;

src/TensorFlowNET.Core/Operations/string_ops.cs

@@ -96,9 +96,9 @@ namespace Tensorflow
                         }
                     }.SetAttributes(new { maxsplit }));
                 var (indices, values, shape) = (result[0], result[1], result[2]);
-                indices.set_shape(new TensorShape(-1, 2));
-                values.set_shape(new TensorShape(-1));
-                shape.set_shape(new TensorShape(2));
+                indices.shape = new Shape(-1, 2);
+                values.shape = new Shape(-1);
+                shape.shape = new Shape(2);
 
                 var sparse_result = new SparseTensor(indices, values, shape);
                 return RaggedTensor.from_value_rowids(sparse_result.values,

src/TensorFlowNET.Core/Operations/weights_broadcast_ops.cs

@@ -29,10 +29,10 @@ namespace Tensorflow
                     weights, dtype: values.dtype.as_base_dtype(), name: "weights");
 
                 // Try static check for exact match.
-                var weights_shape = weights.TensorShape;
-                var values_shape = values.TensorShape;
-                if (weights_shape.is_fully_defined() &&
-                    values_shape.is_fully_defined())
+                var weights_shape = weights.shape;
+                var values_shape = values.shape;
+                if (weights_shape.IsFullyDefined &&
+                    values_shape.IsFullyDefined)
                     return weights;
 
                 return math_ops.multiply(

src/TensorFlowNET.Core/Tensors/Ragged/RaggedTensor.cs

@@ -34,13 +34,13 @@ namespace Tensorflow
         Tensor _row_splits => _row_partition.row_splits;
 
         public TF_DataType dtype => _values.dtype;
-        public TensorShape shape
+        public Shape shape
         {
             get
             {
                 var nrows = _row_partition.static_nrows;
                 var ncols = _row_partition.static_uniform_row_length;
-                return new TensorShape(nrows, ncols);
+                return new Shape(nrows, ncols);
             }
         }
 

src/TensorFlowNET.Core/Tensors/Ragged/SparseTensor.cs

@@ -55,12 +55,12 @@ namespace Tensorflow
 
         void _init()
         {
-            var indices_shape = indices.TensorShape.with_rank(2);
-            var values_shape = values.TensorShape.with_rank(1);
-            var dense_shape_shape = dense_shape.TensorShape.with_rank(1);
+            var indices_shape = indices.shape.with_rank(2);
+            var values_shape = values.shape.with_rank(1);
+            var dense_shape_shape = dense_shape.shape.with_rank(1);
 
-            indices_shape["0"].merge_with(values_shape[0]);
-            indices_shape["1"].merge_with(dense_shape_shape[0]);
+            indices_shape["0"].merge_with(new Shape(values_shape[0]));
+            indices_shape["1"].merge_with(new Shape(dense_shape_shape[0]));
         }
 
         public static implicit operator Tensor(SparseTensor indexedSlices)

src/TensorFlowNET.Core/Tensors/Tensor.Creation.cs

@@ -76,7 +76,7 @@ namespace Tensorflow
         public Tensor(ulong value) => InitTensor(new[] { value }, Shape.Scalar);
         public Tensor(float value) => InitTensor(new[] { value }, Shape.Scalar);
         public Tensor(double value) => InitTensor(new[] { value }, Shape.Scalar);
-        public Tensor(string value) => InitTensor(new[] { value }, TensorShape.Scalar);
+        public Tensor(string value) => InitTensor(new[] { value }, Shape.Scalar);
         #endregion
 
         #region 1d array
@@ -116,7 +116,7 @@ namespace Tensorflow
         protected unsafe void InitTensor(byte[] bytes, TF_DataType dtype)
         {
             if (dtype == TF_DataType.TF_STRING)
-                _handle = StringTensor(new byte[][] { bytes }, TensorShape.Scalar);
+                _handle = StringTensor(new byte[][] { bytes }, Shape.Scalar);
             else
                 throw new NotImplementedException("");
             isCreatedInGraphMode = !tf.executing_eagerly();

src/TensorFlowNET.Core/Tensors/Tensor.Explicit.cs

@@ -138,10 +138,10 @@ namespace Tensorflow
             if (tensor == null)
                 throw new ArgumentNullException(nameof(tensor));
 
-            if (tensor.TensorShape.ndim != 0)
+            if (tensor.shape.ndim != 0)
                 throw new ArgumentException("Tensor must have 0 dimensions in order to convert to scalar");
 
-            if (tensor.TensorShape.size != 1)
+            if (tensor.shape.size != 1)
                 throw new ArgumentException("Tensor must have size 1 in order to convert to scalar");
         }
 

src/TensorFlowNET.Core/Tensors/Tensor.String.cs

@@ -10,7 +10,7 @@ namespace Tensorflow
     {
         const int TF_TSRING_SIZE = 24;
 
-        public IntPtr StringTensor(string[] strings, TensorShape shape)
+        public IntPtr StringTensor(string[] strings, Shape shape)
         {
             // convert string array to byte[][]
             var buffer = new byte[strings.Length][];
@@ -20,7 +20,7 @@ namespace Tensorflow
             return StringTensor(buffer, shape);
         }
 
-        public IntPtr StringTensor(byte[][] buffer, TensorShape shape)
+        public IntPtr StringTensor(byte[][] buffer, Shape shape)
         {
             var handle = c_api.TF_AllocateTensor(TF_DataType.TF_STRING,
                 shape.ndim == 0 ? null : shape.dims,
@@ -63,7 +63,7 @@ namespace Tensorflow
             // [offset1, offset2,...,offsetn, s1size, s1bytes, s2size, s2bytes,...,snsize,snbytes]
             //
             long size = 1;
-            foreach (var s in TensorShape.dims)
+            foreach (var s in shape.dims)
                 size *= s;
 
             var buffer = new byte[size][];

src/TensorFlowNET.Core/Tensors/Tensor.cs

@@ -111,7 +111,10 @@ namespace Tensorflow
         {
             get
             {
-                var dims = new long[rank < 0 ? 0 : rank];
+                if (rank < 0)
+                    return Shape.Null;
+
+                var dims = new Shape(new long[rank]);
 
                 if (_handle == IntPtr.Zero)
                 {
@@ -128,6 +131,13 @@ namespace Tensorflow
 
             set
             {
+                if (this is EagerTensor)
+                {
+                    if(!shape.is_compatible_with(value))
+                        throw new ValueError($"Tensor's shape is not compatible.");
+                    return;
+                }
+
                 if (value == null)
                     c_api.TF_GraphSetTensorShape(graph, _as_tf_output(), null, -1, tf.Status.Handle);
                 else
@@ -142,22 +152,12 @@ namespace Tensorflow
             return rank < 0 ? null : shape.dims.Select(x => (int)x).ToArray();
         }
 
-        public TensorShape TensorShape => rank < 0 ? new TensorShape() : shape;
-
         /// <summary>
         /// Keras History: (Layer, (node_index, tensor_index))
         /// </summary>
         public KerasHistory KerasHistory { get; set; }
         public Tensor KerasMask { get; set; }
 
-        /// <summary>
-        ///     Updates the shape of this tensor.
-        /// </summary>
-        public virtual void set_shape(TensorShape shape)
-        {
-            this.shape = shape.rank >= 0 ? shape : null;
-        }
-
         /// <summary>
         ///     Updates the shape of this tensor.
         /// </summary>
@@ -250,11 +250,11 @@ namespace Tensorflow
             switch (rank)
             {
                 case -1:
-                    return $"tf.Tensor '{name}' shape={TensorShape} dtype={dtype.as_numpy_name()}";
+                    return $"tf.Tensor '{name}' shape={shape} dtype={dtype.as_numpy_name()}";
                 case 0:
-                    return $"tf.Tensor '{name}' shape={TensorShape} dtype={dtype.as_numpy_name()}";
+                    return $"tf.Tensor '{name}' shape={shape} dtype={dtype.as_numpy_name()}";
                 default:
-                    return $"tf.Tensor '{name}' shape={TensorShape} dtype={dtype.as_numpy_name()}";
+                    return $"tf.Tensor '{name}' shape={shape} dtype={dtype.as_numpy_name()}";
             }
         }
 
@@ -263,7 +263,7 @@ namespace Tensorflow
             if (dtype == TF_DataType.TF_STRING)
             {
                 long size = 1;
-                foreach (var s in TensorShape.dims)
+                foreach (var s in shape.dims)
                     size *= s;
                 var tstr = TensorDataPointer;
 

src/TensorFlowNET.Core/Tensors/TensorArray.cs

@@ -37,7 +37,7 @@ namespace Tensorflow
 
         public TensorArray(TF_DataType dtype, Tensor size = default, bool? clear_after_read = null, bool? dynamic_size = null,
             string tensor_array_name = null, Tensor handle = null, Tensor flow = null,
-            bool infer_shape = true, TensorShape element_shape = null,
+            bool infer_shape = true, Shape element_shape = null,
             bool colocate_with_first_write_call = true, string name = null)
         {
             _implementation = new _GraphTensorArray(dtype,

src/TensorFlowNET.Core/Tensors/TensorShape.Convert.cs

@@ -1,47 +0,0 @@
-using System.Linq;
-using Tensorflow.NumPy;
-
-namespace Tensorflow
-{
-    public partial class TensorShape
-    {
-        public void Deconstruct(out long h, out long w)
-        {
-            h = dims[0];
-            w = dims[1];
-        }
-
-        public static implicit operator TensorShape(Shape shape) => new TensorShape((long[])shape.dims.Clone());
-        public static implicit operator Shape(TensorShape shape) => shape == null ? null : new Shape((long[])shape.dims.Clone());
-
-        public static implicit operator int[](TensorShape shape) => shape == null ? null : shape.dims.Select(x => (int)x).ToArray(); //we clone to avoid any changes
-        public static implicit operator TensorShape(int[] dims) => dims == null ? null : new TensorShape(dims);
-
-        public static implicit operator long[](TensorShape shape) => shape == null ? null : (long[])shape.dims.Clone(); //we clone to avoid any changes
-        public static implicit operator TensorShape(long[] dims) => dims == null ? null : new TensorShape(dims);
-
-        public static explicit operator long(TensorShape shape) => shape.size;
-        public static implicit operator TensorShape(long dim) => new TensorShape(dim);
-
-        public static explicit operator (long, long)(TensorShape shape) => shape.dims.Length == 2 ? (shape.dims[0], shape.dims[1]) : (0, 0);
-        public static implicit operator TensorShape((long, long) dims) => new TensorShape(dims.Item1, dims.Item2);
-
-        public static explicit operator (long, long, long)(TensorShape shape) => shape.dims.Length == 3 ? (shape.dims[0], shape.dims[1], shape.dims[2]) : (0, 0, 0);
-        public static implicit operator TensorShape((long, long, long) dims) => new TensorShape(dims.Item1, dims.Item2, dims.Item3);
-
-        public static explicit operator (long, long, long, long)(TensorShape shape) => shape.dims.Length == 4 ? (shape.dims[0], shape.dims[1], shape.dims[2], shape.dims[3]) : (0, 0, 0, 0);
-        public static implicit operator TensorShape((long, long, long, long) dims) => new TensorShape(dims.Item1, dims.Item2, dims.Item3, dims.Item4);
-
-        public static explicit operator (long, long, long, long, long)(TensorShape shape) => shape.dims.Length == 5 ? (shape.dims[0], shape.dims[1], shape.dims[2], shape.dims[3], shape.dims[4]) : (0, 0, 0, 0, 0);
-        public static implicit operator TensorShape((long, long, long, long, long) dims) => new TensorShape(dims.Item1, dims.Item2, dims.Item3, dims.Item4, dims.Item5);
-
-        public static explicit operator (long, long, long, long, long, long)(TensorShape shape) => shape.dims.Length == 6 ? (shape.dims[0], shape.dims[1], shape.dims[2], shape.dims[3], shape.dims[4], shape.dims[5]) : (0, 0, 0, 0, 0, 0);
-        public static implicit operator TensorShape((long, long, long, long, long, long) dims) => new TensorShape(dims.Item1, dims.Item2, dims.Item3, dims.Item4, dims.Item5, dims.Item6);
-
-        public static explicit operator (long, long, long, long, long, long, long)(TensorShape shape) => shape.dims.Length == 7 ? (shape.dims[0], shape.dims[1], shape.dims[2], shape.dims[3], shape.dims[4], shape.dims[5], shape.dims[6]) : (0, 0, 0, 0, 0, 0, 0);
-        public static implicit operator TensorShape((long, long, long, long, long, long, long) dims) => new TensorShape(dims.Item1, dims.Item2, dims.Item3, dims.Item4, dims.Item5, dims.Item6, dims.Item7);
-
-        public static explicit operator (long, long, long, long, long, long, long, long)(TensorShape shape) => shape.dims.Length == 8 ? (shape.dims[0], shape.dims[1], shape.dims[2], shape.dims[3], shape.dims[4], shape.dims[5], shape.dims[6], shape.dims[7]) : (0, 0, 0, 0, 0, 0, 0, 0);
-        public static implicit operator TensorShape((long, long, long, long, long, long, long, long) dims) => new TensorShape(dims.Item1, dims.Item2, dims.Item3, dims.Item4, dims.Item5, dims.Item6, dims.Item7, dims.Item8);
-    }
-}

src/TensorFlowNET.Core/Tensors/TensorShape.Equals.cs

@@ -1,37 +0,0 @@
-using System;
-using System.Linq;
-
-namespace Tensorflow
-{
-    public partial class TensorShape
-    {
-        public override bool Equals(Object obj)
-        {
-            switch (obj)
-            {
-                case TensorShape shape1:
-                    if (rank == -1 && shape1.rank == -1)
-                        return false;
-                    else if (rank != shape1.rank)
-                        return false;
-                    return Enumerable.SequenceEqual(shape1.dims, dims);
-                case long[] shape2:
-                    if (rank != shape2.Length)
-                        return false;
-                    return Enumerable.SequenceEqual(dims, shape2);
-                default:
-                    return false;
-            }
-        }
-
-        /*public static bool operator ==(TensorShape shape1, TensorShape shape2)
-        {
-            return false;
-        }
-
-        public static bool operator !=(TensorShape shape1, TensorShape shape2)
-        {
-            return false;
-        }*/
-    }
-}

src/TensorFlowNET.Core/Tensors/TensorShape.cs

@@ -1,306 +0,0 @@
-using Tensorflow.NumPy;
-using System;
-using System.Collections.Generic;
-using System.Diagnostics.CodeAnalysis;
-using System.Linq;
-using System.Runtime.CompilerServices;
-using static Tensorflow.Binding;
-
-namespace Tensorflow
-{
-    /// <summary>
-    ///     Represents the shape of a `Tensor`.
-    /// </summary>
-    /// <remarks>https://www.tensorflow.org/api_docs/python/tf/TensorShape</remarks>
-    public partial class TensorShape
-    {
-        private readonly Shape shape;
-
-        /// <summary>
-        ///     Returns a list of Dimensions, or None if the shape is unspecified.
-        /// </summary>
-        public long[] dims => shape.dims;
-
-        /// <summary>
-        ///     Returns the rank of this shape.
-        /// </summary>
-        public int ndim => rank;
-
-        private int _rank;
-        /// <summary>
-        ///     Returns the rank of this shape.
-        /// </summary>
-        public int rank => _rank > -1 ? shape.ndim : -1;
-
-        /// <summary>
-        ///     Returns the size this shape represents.
-        /// </summary>
-        public long size
-        {
-            get
-            {
-                var dims = shape.dims;
-                var computed = 1L;
-                for (int i = 0; i < dims.Length; i++)
-                {
-                    var val = dims[i];
-                    if (val <= 0)
-                        continue;
-                    computed *= val;
-                }
-
-                return computed;
-            }
-        }
-
-        public TensorShape()
-        {
-            _rank = -1;
-            shape = new Shape();
-        }
-
-        public static TensorShape Scalar
-            => new TensorShape(new long[0]);
-
-        public TensorShape(TensorShapeProto proto)
-        {
-            if (proto.UnknownRank) return;
-            switch (proto.Dim.Count)
-            {
-                case 0: shape = new Shape(new long[0]); 
-                    break;
-                default:
-                    var protodims = proto.Dim;
-                    var len = protodims.Count;
-                    var dims = new long[len];
-                    for (int i = 0; i < len; i++)
-                        dims[i] = protodims[i].Size;
-                    shape = new Shape(dims); 
-                    break;
-            }
-        }
-
-        public TensorShape(params int[] dims)
-        {
-            switch (dims.Length)
-            {
-                case 0:
-                    shape = new Shape(new long[0]);
-                    break;
-                default:
-                    shape = new Shape(dims.Select(x => Convert.ToInt64(x)).ToArray());
-                    break;
-            }
-        }
-
-        public TensorShape(params long[] dims)
-        {
-            switch (dims.Length)
-            {
-                case 0: shape = new Shape(new long[0]); 
-                    break;
-                default: shape = new Shape(dims); 
-                    break;
-            }
-        }
-
-        public TensorShape(long[][] dims)
-        {
-            if (dims.Length == 1)
-            {
-                switch (dims[0].Length)
-                {
-                    case 0: shape = new Shape(new long[0]); 
-                        break;
-                    default: shape = new Shape(dims[0]); 
-                        break;
-                }
-            }
-            else
-            {
-                throw new NotImplementedException("TensorShape int[][] dims");
-            }
-        }
-
-        /// <summary>
-        /// 
-        /// </summary>
-        /// <param name="slice"></param>
-        /// <returns></returns>
-        /// <exception cref="ArgumentException">When <see cref="Slice"/> is not an Index.</exception>
-        [SuppressMessage("ReSharper", "PossibleInvalidOperationException")]
-        public TensorShape this[Slice slice]
-        {
-            get
-            {
-                if (!slice.Stop.HasValue)
-                    slice.Stop = dims.Length - slice.Start + 1;
-
-                if (slice.Start.HasValue == false || slice.Length.HasValue == false)
-                    throw new ArgumentException("Slice must has Start and Length.");
-
-                return new TensorShape(dims.Skip(slice.Start.Value)
-                    .Take(slice.Length.Value)
-                    .ToArray());
-            }
-        }
-
-        public long this[int index] => index < 0 ? dims[ndim + index] : dims[index];
-
-        /// <summary>
-        ///     Returns True iff `self` is fully defined in every dimension.
-        /// </summary>
-        /// <returns></returns>
-        public bool is_fully_defined()
-        {
-            return rank > -1 && dims != null && dims.Count(x => x < 1) == 0;
-        }
-
-        public bool is_compatible_with(TensorShape shape2)
-        {
-            if (dims != null && shape2.dims != null)
-            {
-                if (dims.Contains(-1) || shape2.dims.Contains(-1))
-                    return true;
-
-                if (shape.size != (ulong)shape2.size)
-                    return false;
-            }
-
-            return true;
-        }
-
-        public void assert_has_rank(int rank)
-        {
-            if (rank != ndim)
-                throw new ValueError(String.Format("Shape {0} must have rank {1}", ndim, rank));
-        }
-
-        [SuppressMessage("ReSharper", "ParameterHidesMember")]
-        public TensorShape with_rank_at_least(int rank)
-        {
-            if (ndim < rank)
-                throw new ValueError($"Shape {this} must have rank at least {rank}");
-            else
-                return this;
-        }
-
-        public TensorShape with_rank(int rank)
-        {
-            return merge_with(unknown_shape(rank: rank));
-        }
-
-        /// <summary>
-        /// Returns an unknown TensorShape, optionally with a known rank.
-        /// </summary>
-        /// <param name="rank"></param>
-        /// <returns></returns>
-        public TensorShape unknown_shape(int rank = -1)
-        {
-            if (rank == -1)
-                return new TensorShape(-1);
-            else
-                return new TensorShape(Enumerable.Repeat(-1L, rank).ToArray());
-        }
-
-        /// <summary>
-        ///     Returns the concatenation of the dimension in `self` and `other`.
-        /// </summary>
-        /// <param name="other"></param>
-        /// <returns></returns>
-        [MethodImpl(MethodImplOptions.AggressiveInlining)]
-        public TensorShape concatenate(long[] other)
-        {
-            return concatenate(new TensorShape(other));
-        }
-
-        /// <summary>
-        ///     Returns the concatenation of the dimension in `self` and `other`.
-        /// </summary>
-        /// <param name="other"></param>
-        /// <returns></returns>
-        public TensorShape concatenate(TensorShape other)
-        {
-            var otherShape = other;
-
-            if (ndim < 0 || otherShape.ndim < 0)
-                return new TensorShape();
-            else
-            {
-                var concatenate_dims = new long[ndim + otherShape.ndim];
-                for (int i = 0; i < ndim; i++)
-                    concatenate_dims[i] = dims[i];
-
-                for (int i = 0; i < otherShape.ndim; i++)
-                    concatenate_dims[ndim + i] = otherShape.dims[i];
-
-                return new TensorShape(concatenate_dims);
-            }
-        }
-
-        /// <summary>
-        /// Returns a `TensorShape` combining the information in `self` and `other`.
-        /// </summary>
-        /// <param name="other"></param>
-        /// <returns></returns>
-        public TensorShape merge_with(TensorShape other)
-        {
-            if (dims == null)
-                return other;
-
-            var new_dims = new List<long>();
-
-            foreach (var i in range(ndim))
-            {
-                var dim = new Dimension(dims[i]);
-                var merged = dim.merge_with(new Dimension(other.dims[i]));
-                new_dims.Add(merged.value);
-            }
-
-            return new TensorShape(new_dims.ToArray());
-        }
-
-        /// <summary>
-        ///     Returns a cloned array from <see cref="dims"/>.
-        /// </summary>
-        public long[] as_list()
-        {
-            if (shape.IsEmpty)
-                throw new ValueError("as_list() is not defined on an unknown TensorShape.");
-            return (long[])dims.Clone();
-        }
-
-        public long[] as_list_long()
-        {
-            if (shape.IsEmpty)
-                throw new ValueError("as_list() is not defined on an unknown TensorShape.");
-            return dims.Select(x => Convert.ToInt64(x)).ToArray();
-        }
-
-        public long num_elements()
-        {
-            if (is_fully_defined())
-            {
-                var size = 1L;
-                foreach (var dim in dims)
-                    size *= dim;
-                return size;
-            }
-
-            return -1;
-        }
-
-        public override string ToString()
-        {
-            switch (rank)
-            {
-                case -1:
-                    return $"<unknown>";
-                case 0:
-                    return $"()";
-                default:
-                    return $"{string.Join(",", shape).Replace("-1", "None")}";
-            }
-        }
-    }
-}

src/TensorFlowNET.Core/Tensors/Tensors.cs

@@ -18,7 +18,7 @@ namespace Tensorflow
         List<Tensor> items = new List<Tensor>();
 
         public TF_DataType dtype => items.First().dtype;
-        public TensorShape shape => items.First().TensorShape;
+        public Shape shape => items.First().shape;
         public int rank => items.First().rank;
         public Graph graph => items.First().graph;
         public bool IsCreatedInGraphMode => items.First().IsCreatedInGraphMode;