_CastGrad

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

@@ -37,15 +37,8 @@ namespace Tensorflow.Eager
             => NewEagerTensorHandle(_handle); 
         #endregion
 
-        public EagerTensor(object value,string device_name, TF_DataType dtype = TF_DataType.TF_UINT8) : base((float[])value)
-        {
-            throw new NotImplementedException("");
-        }
-
         public EagerTensor(object value, Shape? shape = null, string device_name = null, TF_DataType dtype = TF_DataType.TF_UINT8) : base((float[])value)
-        {
-            NewEagerTensorHandle(_handle);
-        }
+            => NewEagerTensorHandle(_handle);
 
         public EagerTensor(Shape shape, TF_DataType dtype) : base(shape, dtype)
             => NewEagerTensorHandle(_handle);
@@ -56,6 +49,9 @@ namespace Tensorflow.Eager
         public EagerTensor(byte[] bytes, Shape shape, TF_DataType dtype) : base(bytes, shape, dtype)
             => NewEagerTensorHandle(_handle);
 
+        public EagerTensor(IntPtr data_ptr, Shape shape, TF_DataType dtype) : base(data_ptr, shape, dtype)
+            => NewEagerTensorHandle(_handle);
+
         void NewEagerTensorHandle(SafeTensorHandle h)
         {
             _id = ops.uid();

src/TensorFlowNET.Core/Gradients/image_grad.cs

@@ -30,7 +30,7 @@ namespace Tensorflow.Gradients
             var shape = new Shape(image.shape.dims.Skip(1).Take(2).ToArray());
             Tensor image_shape = null;
             if (shape.IsFullyDefined)
-                image_shape = constant_op.constant(image.shape.dims.Skip(1).Take(2).ToArray());
+                image_shape = constant_op.constant(image.shape.as_int_list().Skip(1).Take(2).ToArray());
             else
                 image_shape = array_ops.shape(image)["1:3"];
 

src/TensorFlowNET.Core/Gradients/math_grad.cs

@@ -694,6 +694,20 @@ namespace Tensorflow.Gradients
             });
         }
 
+        [RegisterGradient("Cast")]
+        public static Tensor[] _CastGrad(Operation op, Tensor[] grads)
+        {
+            var grad = grads[0];
+            var x = op.inputs[0];
+
+            var src_type = x.dtype.as_base_dtype();
+            var dst_type = grad.dtype.as_base_dtype();
+            if (src_type.is_value_dtype() && dst_type.is_value_dtype())
+                return new Tensor[] { math_ops.cast(grad, src_type) };
+            else
+                return new Tensor[0];
+        }
+
         [RegisterGradient("Cos")]
         public static Tensor[] _CosGrad(Operation op, Tensor[] grads)
         {

src/TensorFlowNET.Core/NumPy/Numpy.Math.cs

@@ -13,7 +13,7 @@ namespace Tensorflow.NumPy
         public static NDArray exp(NDArray x) => new NDArray(tf.exp(x));
 
         [AutoNumPy]
-        public static NDArray floor(NDArray x) => new NDArray(tf.floor(x));
+        public static NDArray floor(NDArray x) => new NDArray(math_ops.floor(x));
 
         [AutoNumPy]
         public static NDArray log(NDArray x) => new NDArray(tf.log(x));

src/TensorFlowNET.Core/Operations/math_ops.cs

@@ -720,6 +720,8 @@ namespace Tensorflow
                 return gen_math_ops.range(start1, limit1, delta1, name);
             });
         }
+        public static Tensor floor(Tensor x, string name = null)
+            => tf.Context.ExecuteOp("Floor", name, new ExecuteOpArgs(x));
 
         public static Tensor floordiv(Tensor x, Tensor y, string name = null)
         {

src/TensorFlowNET.Core/Tensorflow.Binding.csproj

@@ -5,7 +5,7 @@
     <AssemblyName>TensorFlow.NET</AssemblyName>
     <RootNamespace>Tensorflow</RootNamespace>
     <TargetTensorFlow>2.2.0</TargetTensorFlow>
-    <Version>0.60.0</Version>
+    <Version>0.60.1</Version>
     <LangVersion>9.0</LangVersion>
     <Nullable>enable</Nullable>
     <Authors>Haiping Chen, Meinrad Recheis, Eli Belash</Authors>
@@ -20,8 +20,8 @@
     <Description>Google's TensorFlow full binding in .NET Standard.
 Building, training and infering deep learning models.
 https://tensorflownet.readthedocs.io</Description>
-    <AssemblyVersion>0.60.0.0</AssemblyVersion>
-    <PackageReleaseNotes>tf.net 0.20.x and above are based on tensorflow native 2.x.
+    <AssemblyVersion>0.60.1.0</AssemblyVersion>
+    <PackageReleaseNotes>tf.net 0.60.x and above are based on tensorflow native 2.6.0
 
 * Eager Mode is added finally.
 * tf.keras is partially working. 
@@ -35,7 +35,7 @@ Keras API is a separate package released as TensorFlow.Keras.
 tf.net 0.4x.x aligns with TensorFlow v2.4.1 native library.
 tf.net 0.5x.x aligns with TensorFlow v2.5.x native library.
 tf.net 0.6x.x aligns with TensorFlow v2.6.x native library.</PackageReleaseNotes>
-    <FileVersion>0.60.0.0</FileVersion>
+    <FileVersion>0.60.1.0</FileVersion>
     <PackageLicenseFile>LICENSE</PackageLicenseFile>
     <PackageRequireLicenseAcceptance>true</PackageRequireLicenseAcceptance>
     <SignAssembly>true</SignAssembly>

src/TensorFlowNET.Core/Tensors/Tensors.cs

@@ -79,6 +79,7 @@ namespace Tensorflow
         public static implicit operator Tensors((Tensor, Tensor) tuple)
             => new Tensors(tuple.Item1, tuple.Item2);
 
+        [AutoNumPy]
         public static implicit operator Tensors(NDArray nd)
             => new Tensors(nd);
 

src/TensorFlowNET.Core/Tensors/dtypes.cs

@@ -138,7 +138,8 @@ namespace Tensorflow
                     dtype = TF_DataType.TF_BOOL;
                     break;
                 default:
-                    throw new NotSupportedException($"Unable to convert {type} to a TensorFlow data type.");
+                    dtype = TF_DataType.DtInvalid;
+                    break;
             }
 
             return dtype;

src/TensorFlowNET.Core/ops.cs

@@ -165,7 +165,7 @@ namespace Tensorflow
             if (dtype == TF_DataType.TF_STRING)
                 return ret;
 
-            if (dtype.as_base_dtype() != ret.dtype.as_base_dtype())
+            if (dtype != TF_DataType.DtInvalid && dtype.as_base_dtype() != ret.dtype.as_base_dtype())
                 ret = gen_math_ops.cast(ret, dtype, name: name);
 
             return ret;

src/TensorFlowNET.Keras/Engine/Model.Training.cs

@@ -4,6 +4,7 @@ using System.Text;
 using HDF.PInvoke;
 using HDF5CSharp;
 using Tensorflow.NumPy;
+using static Tensorflow.Binding;
 using Tensorflow.Keras.Saving;
 
 namespace Tensorflow.Keras.Engine
@@ -14,7 +15,11 @@ namespace Tensorflow.Keras.Engine
         public void load_weights(string filepath, bool by_name = false, bool skip_mismatch = false, object options = null)
         {
             long fileId = Hdf5.OpenFile(filepath, true);
-
+            if(fileId < 0)
+            {
+                tf_output_redirect.WriteLine($"Can't find weights file {filepath}");
+                return;
+            }
             bool msuccess = Hdf5.GroupExists(fileId, "model_weights");
             bool lsuccess = Hdf5.GroupExists(fileId, "layer_names");
 

src/TensorFlowNET.Keras/Preprocessings/Preprocessing.paths_and_labels_to_dataset.cs

@@ -26,12 +26,12 @@ namespace Tensorflow.Keras
                 images[i] = resized_image.numpy();
                 tf_output_redirect.WriteLine(image_paths[i]);
             };
+            var img_ds = tf.data.Dataset.from_tensor_slices(images);
 
             // option 2: dynamic load, but has error, need to fix
-            /* var path_ds = tf.data.Dataset.from_tensor_slices(image_paths);
-            var img_ds = path_ds.map(x => path_to_image(x, image_size, num_channels, interpolation));*/
-
-            var img_ds = tf.data.Dataset.from_tensor_slices(images);
+            // var path_ds = tf.data.Dataset.from_tensor_slices(image_paths);
+            // var img_ds = path_ds.map(x => path_to_image(x, image_size, num_channels, interpolation));
+            
             if (label_mode == "int")
             {
                 var label_ds = dataset_utils.labels_to_dataset(labels, label_mode, num_classes);
@@ -43,6 +43,7 @@ namespace Tensorflow.Keras
 
         Tensor path_to_image(Tensor path, Shape image_size, int num_channels, string interpolation)
         {
+            tf.print(path);
             var img = tf.io.read_file(path);
             img = tf.image.decode_image(
                 img, channels: num_channels, expand_animations: false);
@@ -57,8 +58,18 @@ namespace Tensorflow.Keras
             int num_classes,
             int max_length = -1)
         {
-            var path_ds = tf.data.Dataset.from_tensor_slices(image_paths);
-            var string_ds = path_ds.map(x => path_to_string_content(x, max_length));
+            var text = new string[image_paths.Length];
+            for (int i = 0; i < text.Length; i++)
+            {
+                text[i] = File.ReadAllText(image_paths[i]);
+                tf_output_redirect.WriteLine(image_paths[i]);
+            }
+
+            var images = np.array(text);
+            var string_ds = tf.data.Dataset.from_tensor_slices(images);
+
+            // var path_ds = tf.data.Dataset.from_tensor_slices(image_paths);
+            // var string_ds = path_ds.map(x => path_to_string_content(x, max_length));
 
             if (label_mode == "int")
             {

src/TensorFlowNET.Keras/Tensorflow.Keras.csproj

@@ -7,7 +7,7 @@
     <Nullable>enable</Nullable>
     <RootNamespace>Tensorflow.Keras</RootNamespace>
     <Platforms>AnyCPU;x64</Platforms>
-    <Version>0.6.0</Version>
+    <Version>0.6.1</Version>
     <Authors>Haiping Chen</Authors>
     <Product>Keras for .NET</Product>
     <Copyright>Apache 2.0, Haiping Chen 2021</Copyright>
@@ -37,8 +37,8 @@ Keras is an API designed for human beings, not machines. Keras follows best prac
     <RepositoryType>Git</RepositoryType>
     <SignAssembly>true</SignAssembly>
     <AssemblyOriginatorKeyFile>Open.snk</AssemblyOriginatorKeyFile>
-    <AssemblyVersion>0.6.0.0</AssemblyVersion>
-    <FileVersion>0.6.0.0</FileVersion>
+    <AssemblyVersion>0.6.1.0</AssemblyVersion>
+    <FileVersion>0.6.1.0</FileVersion>
     <PackageLicenseFile>LICENSE</PackageLicenseFile>
   </PropertyGroup>