fix keras reshape layer.

src/TensorFlowNET.Core/Binding.Util.cs

@@ -540,8 +540,10 @@ namespace Tensorflow
                     return nd.dtype;
                 case Tensor tensor:
                     return tensor.dtype;
-                case Tensor[] tensor:
-                    return tensor[0].dtype;
+                case Tensors tensors:
+                    return tensors.dtype;
+                case IEnumerable<Tensor> tensors:
+                    return tensors.First().dtype;
                 case ResourceVariable variable:
                     return variable.dtype;
                 default:

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

@@ -51,7 +51,8 @@ namespace Tensorflow.Eager
                     SafeTensorHandleHandle tensor_handle = inputs[i] switch
                     {
                         EagerTensor et => et.EagerTensorHandle,
-                        _ => throw new NotImplementedException("")
+                        Tensor nd => nd.EagerTensorHandle,
+                        _ => throw new NotImplementedException("Eager tensor handle has not been allocated.")
                     };
                     c_api.TFE_OpAddInput(op, tensor_handle, status.Handle);
                     status.Check(true);

src/TensorFlowNET.Core/Operations/array_ops.cs

@@ -186,7 +186,10 @@ namespace Tensorflow
 
         private static Tensor _constant_if_small(int value, Tensor shape)
         {
-            return shape < 1000L;
+            if (shape.dtype == TF_DataType.TF_INT64)
+                return shape < 1000L;
+            else
+                return shape < 1000;
         }
 
         private static Tensor _constant_if_small<T>(T value, Shape shape, TF_DataType dtype, string name)

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

@@ -16,7 +16,7 @@ namespace Tensorflow
         {
             //Are the types matching?
             if (typeof(T).as_tf_dtype() != dtype)
-                throw new ArrayTypeMismatchException($"dtype {dtype} mismatch.");
+                throw new ArrayTypeMismatchException($"Required dtype {dtype} mismatch with {typeof(T).as_tf_dtype()}.");
 
             if (ndim == 0 && size == 1)  //is it a scalar?
             {

src/TensorFlowNET.Core/Tensors/constant_op.cs

@@ -74,7 +74,7 @@ namespace Tensorflow
             if (dtype != TF_DataType.DtInvalid &&
                 value.GetType().Name != "NDArray" &&
                 value.GetType().BaseType.Name != "Array" &&
-                dtypes.as_base_dtype(dtype) != dtypes.as_tf_dtype(value.GetType()))
+                dtype != value.GetDataType())
             {
                 switch (dtype)
                 {
@@ -87,6 +87,9 @@ namespace Tensorflow
                     case TF_DataType.TF_INT64:
                         value = Convert.ToInt64(value);
                         break;
+                    case TF_DataType.TF_INT32:
+                        value = Convert.ToInt32(value);
+                        break;
                     default:
                         break;
                 }

src/TensorFlowNET.Core/ops.cs

@@ -162,7 +162,7 @@ namespace Tensorflow
                             ? tensor.AsPlaceholder(name: name)
                             : tensor.AsConstant(name: name),
                 Tensor tensor => tensor,
-                Tensor[] tensors => array_ops._autopacking_helper(tensors, dtype, name == null ? "packed" : name),
+                IEnumerable<Tensor> tensors => array_ops._autopacking_helper(tensors, dtype, name == null ? "packed" : name),
                 RefVariable varVal => varVal._TensorConversionFunction(dtype: dtype, name: name, as_ref: as_ref),
                 ResourceVariable varVal => varVal._TensorConversionFunction(dtype: dtype, name: name, as_ref: as_ref),
                 Axis ts => constant_op.constant(ts.axis, dtype: dtype, name: name),

src/TensorFlowNET.Keras/Layers/Reshaping/Reshape.cs

@@ -21,12 +21,14 @@ namespace Tensorflow.Keras.Layers
 
         protected override Tensors Call(Tensors inputs, Tensor state = null, bool? training = null)
         {
-            var shapes = new List<object>();
+            var shapes = new List<Tensor>();
             shapes.Add(array_ops.shape(inputs)[0]);
+            var dtype = shapes[0].dtype;
             if (args.TargetShapeObjects != null)
-                shapes.AddRange(args.TargetShapeObjects);
+                // shapes.AddRange(args.TargetShapeObjects);
+                throw new NotImplementedException("");
             if (args.TargetShape != null)
-                args.TargetShape.dims.ToList().ForEach(x => shapes.Add(x));
+                shapes.AddRange(args.TargetShape.dims.Select(x => constant_op.constant(x, dtype)));
             var shape = ops.convert_to_tensor(shapes);
 
             var result = array_ops.reshape(inputs, shape);

src/TensorFlowNET.Keras/Utils/layer_utils.cs

@@ -184,7 +184,7 @@ namespace Tensorflow.Keras.Utils
         public static int count_params(Layer layer, List<IVariableV1> weights)
         {
             var weight_shapes = weights.Select(x => x.shape).ToArray();
-            var total = weight_shapes.Select(p => (int)np.prod(p.dims)).Sum();
+            var total = weight_shapes.Select(p => (int)p.size).Sum();
             return total;
         }
 

test/TensorFlowNET.Keras.UnitTest/Layers/Layers.Reshaping.Test.cs

@@ -1,4 +1,5 @@
 using Microsoft.VisualStudio.TestTools.UnitTesting;
+using Tensorflow;
 using Tensorflow.NumPy;
 using static Tensorflow.Binding;
 using static Tensorflow.KerasApi;
@@ -11,8 +12,8 @@ namespace TensorFlowNET.Keras.UnitTest
         [TestMethod]
         public void ZeroPadding2D()
         {
-            var input_shape = new[] { 1, 1, 2, 2 };
-            var x = np.arange((int)np.prod(input_shape)).reshape(input_shape);
+            Shape input_shape = (1, 1, 2, 2);
+            var x = np.arange(input_shape.size).reshape(input_shape);
             var zero_padding_2d = keras.layers.ZeroPadding2D(new[,] { { 1, 0 }, { 1, 0 } });
             var y = zero_padding_2d.Apply(x);
             Assert.AreEqual((1, 2, 3, 2), y.shape);
@@ -21,8 +22,8 @@ namespace TensorFlowNET.Keras.UnitTest
         [TestMethod]
         public void UpSampling2D()
         {
-            var input_shape = new[] { 2, 2, 1, 3 };
-            var x = np.arange((int)np.prod(input_shape)).reshape(input_shape);
+            Shape input_shape = (2, 2, 1, 3);
+            var x = np.arange(input_shape.size).reshape(input_shape);
             var y = keras.layers.UpSampling2D(size: (1, 2)).Apply(x);
             Assert.AreEqual((2, 2, 2, 3), y.shape);
         }

test/TensorFlowNET.Keras.UnitTest/Layers/MeanSquaredError.Test.cs

@@ -21,7 +21,7 @@ namespace TensorFlowNET.Keras.UnitTest
         {
             var mse = keras.losses.MeanSquaredError();
             var call = mse.Call(y_true, y_pred);
-            Assert.AreEqual((NDArray)0.5, call.numpy()) ;
+            Assert.AreEqual(call.numpy(), 0.5);
         }
 
         [TestMethod]
@@ -33,7 +33,7 @@ namespace TensorFlowNET.Keras.UnitTest
 
             var mse = keras.losses.MeanSquaredError();
             var call = mse.Call(y_true_float, y_pred_float);
-            Assert.AreEqual((NDArray)0.5, call.numpy());
+            Assert.AreEqual(call.numpy(), 0.5f);
         }
 
         [TestMethod]
@@ -42,7 +42,7 @@ namespace TensorFlowNET.Keras.UnitTest
         {
             var mse = keras.losses.MeanSquaredError();
             var call = mse.Call(y_true, y_pred, sample_weight: (NDArray)new double[] { 0.7, 0.3 });
-            Assert.AreEqual((NDArray)0.25, call.numpy());
+            Assert.AreEqual(call.numpy(), 0.25);
         }
 
         [TestMethod]
@@ -50,7 +50,7 @@ namespace TensorFlowNET.Keras.UnitTest
         {
             var mse = keras.losses.MeanSquaredError(reduction: Reduction.SUM);
             var call = mse.Call(y_true, y_pred);
-            Assert.AreEqual((NDArray)1.0, call.numpy());
+            Assert.AreEqual(call.numpy(), 1.0);
         }
 
         [TestMethod]
@@ -59,7 +59,7 @@ namespace TensorFlowNET.Keras.UnitTest
         {
             var mse = keras.losses.MeanSquaredError(reduction: Reduction.NONE);
             var call = mse.Call(y_true, y_pred);
-            Assert.AreEqual((NDArray)new double[] { 0.5, 0.5 }, call.numpy());
+            Assert.AreEqual(call.numpy(), new double[] { 0.5, 0.5 });
         }
     }
 }

test/TensorFlowNET.Native.UnitTest/Tensors/TensorTest.cs

@@ -67,7 +67,7 @@ namespace Tensorflow.Native.UnitTest.Tensors
             Tensor t = c_api.TF_AllocateTensor(TF_DataType.TF_FLOAT, dims, 2, num_bytes);
             EXPECT_EQ(TF_DataType.TF_FLOAT, t.dtype);
             EXPECT_EQ(2, t.ndim);
-            EXPECT_EQ((int)dims[0], t.shape[0]);
+            EXPECT_EQ(dims[0], t.shape[0]);
             EXPECT_EQ(num_bytes, t.bytesize);
             t.Dispose();
         }
@@ -101,9 +101,9 @@ namespace Tensorflow.Native.UnitTest.Tensors
 
             EXPECT_EQ(tensor.dtype, TF_DataType.TF_FLOAT);
             EXPECT_EQ(tensor.rank, nd.ndim);
-            EXPECT_EQ((int)tensor.shape[0], nd.dims[0]);
-            EXPECT_EQ((int)tensor.shape[1], nd.dims[1]);
-            EXPECT_EQ(tensor.bytesize, (ulong)nd.size * sizeof(float));
+            EXPECT_EQ(tensor.shape[0], nd.dims[0]);
+            EXPECT_EQ(tensor.shape[1], nd.dims[1]);
+            EXPECT_EQ(tensor.bytesize, nd.size * sizeof(float));
             Assert.IsTrue(Enumerable.SequenceEqual(nd.Data<float>(), new float[] { 1, 2, 3, 4, 5, 6 }));
         }