1st stage to integrate with NumSharp.

TensorFlow.NET.sln

@@ -17,7 +17,9 @@ Project("{9A19103F-16F7-4668-BE54-9A1E7A4F7556}") = "TensorFlowBenchmark", "src\
 EndProject
 Project("{9A19103F-16F7-4668-BE54-9A1E7A4F7556}") = "TensorFlowHub", "src\TensorFlowHub\TensorFlowHub.csproj", "{8FD59A5A-97EB-457E-B9F1-D88B0C822C6E}"
 EndProject
-Project("{FAE04EC0-301F-11D3-BF4B-00C04F79EFBC}") = "TensorFlowText", "src\TensorFlowText\TensorFlowText.csproj", "{B598E5D5-BD2D-4191-8532-F2FBAC31AB81}"
+Project("{9A19103F-16F7-4668-BE54-9A1E7A4F7556}") = "TensorFlowText", "src\TensorFlowText\TensorFlowText.csproj", "{B598E5D5-BD2D-4191-8532-F2FBAC31AB81}"
+EndProject
+Project("{9A19103F-16F7-4668-BE54-9A1E7A4F7556}") = "NumSharp.Core", "..\NumSharp\src\NumSharp.Core\NumSharp.Core.csproj", "{F219B9B9-B873-4342-BF85-7E89E7F5D64F}"
 EndProject
 Global
 	GlobalSection(SolutionConfigurationPlatforms) = preSolution
@@ -57,6 +59,10 @@ Global
 		{B598E5D5-BD2D-4191-8532-F2FBAC31AB81}.Debug|Any CPU.Build.0 = Debug|Any CPU
 		{B598E5D5-BD2D-4191-8532-F2FBAC31AB81}.Release|Any CPU.ActiveCfg = Release|Any CPU
 		{B598E5D5-BD2D-4191-8532-F2FBAC31AB81}.Release|Any CPU.Build.0 = Release|Any CPU
+		{F219B9B9-B873-4342-BF85-7E89E7F5D64F}.Debug|Any CPU.ActiveCfg = Debug|Any CPU
+		{F219B9B9-B873-4342-BF85-7E89E7F5D64F}.Debug|Any CPU.Build.0 = Debug|Any CPU
+		{F219B9B9-B873-4342-BF85-7E89E7F5D64F}.Release|Any CPU.ActiveCfg = Release|Any CPU
+		{F219B9B9-B873-4342-BF85-7E89E7F5D64F}.Release|Any CPU.Build.0 = Release|Any CPU
 	EndGlobalSection
 	GlobalSection(SolutionProperties) = preSolution
 		HideSolutionNode = FALSE

src/KerasNET.Core/Layers/Dense.cs

@@ -41,7 +41,7 @@ namespace Keras.Layers
             Console.WriteLine("Building Layer \"" + name + "\" ...");
             if (stddev == -1)
                 stddev = (float)(1 / Math.Sqrt(2));
-            var dim = input_shape.Dimensions;
+            var dim = input_shape.dims;
             var input_dim = dim[dim.Length - 1];
             W = tf.Variable(create_tensor(new int[] { input_dim, units }, seed: seed, stddev: (float)stddev));
             WShape = new TensorShape(W.shape);
@@ -52,7 +52,7 @@ namespace Keras.Layers
             var dot = tf.matmul(x, W);
             if (this.activation != null)
                 dot = activation.Activate(dot);
-            Console.WriteLine("Calling Layer \"" + name + "(" + np.array(dot.TensorShape.Dimensions).ToString() + ")\" ...");
+            Console.WriteLine("Calling Layer \"" + name + "(" + np.array(dot.TensorShape.dims).ToString() + ")\" ...");
             return dot;
         }
         public TensorShape __shape__()
@@ -61,7 +61,7 @@ namespace Keras.Layers
         }
         public TensorShape output_shape(TensorShape input_shape)
         {
-            var output_shape = input_shape.Dimensions;
+            var output_shape = input_shape.dims;
             output_shape[output_shape.Length - 1] = units;
             return new TensorShape(output_shape);
         }

src/TensorFlowNET.Core/Contrib/Learn/Estimators/tensor_signature.cs

@@ -11,13 +11,13 @@ namespace Tensorflow.Contrib.Learn.Estimators
             bool _shape_is_compatible_0dim(Shape _this, Shape _other)
             {
                 var __other = tensor_shape.as_shape(_other);
-                if (_this.Dimensions == null || __other.Dimensions == null)
+                if (_this.Dimensions == null || __other.dims == null)
                     return true;
 
-                if (_this.NDim != __other.NDim)
+                if (_this.NDim != __other.ndim)
                     return false;
 
-                foreach (var (x_dim, y_dim) in _this.Dimensions.Zip(__other.Dimensions, (x_dim, y_dim) => (x_dim, y_dim)))
+                foreach (var (x_dim, y_dim) in _this.Dimensions.Zip(__other.dims, (x_dim, y_dim) => (x_dim, y_dim)))
                 {
                     if (x_dim != y_dim)
                         return false;

src/TensorFlowNET.Core/Framework/tensor_shape.cs

@@ -25,10 +25,6 @@ namespace Tensorflow.Framework
         }
 
         public static TensorShape as_shape(this Shape shape)
-        {
-            if (shape is TensorShape tshape)
-                return tshape;
-            return new TensorShape(shape);
-        }
+             => new TensorShape(shape.Dimensions);
     }
 }

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

@@ -81,7 +81,7 @@ namespace Tensorflow.Keras.Layers
 
         protected override void build(TensorShape input_shape)
         {
-            var ndims = input_shape.NDim;
+            var ndims = input_shape.ndim;
             foreach (var (idx, x) in Python.enumerate(axis))
                 if (x < 0)
                     axis[idx] = ndims + x;
@@ -91,7 +91,7 @@ namespace Tensorflow.Keras.Layers
                     _data_format = "NHWC";
 
             var param_dtype = _dtype == TF_DataType.DtInvalid ? TF_DataType.TF_FLOAT : _dtype;
-            var param_shape = new int[] { input_shape.Dimensions[axis[0]] };
+            var param_shape = new int[] { input_shape.dims[axis[0]] };
 
             if (scale)
                 gamma = add_weight("gamma",

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

@@ -72,8 +72,8 @@ namespace Tensorflow.Keras.Layers
         {
             int channel_axis = data_format == "channels_first" ? 1 : -1;
             int input_dim = channel_axis < 0 ? 
-                input_shape.Dimensions[input_shape.NDim + channel_axis] : 
-                input_shape.Dimensions[channel_axis];
+                input_shape.dims[input_shape.ndim + channel_axis] : 
+                input_shape.dims[channel_axis];
             var kernel_shape = new int[] { kernel_size[0], kernel_size[1], input_dim, filters };
             kernel = add_weight(name: "kernel",
                 shape: kernel_shape,

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

@@ -51,7 +51,7 @@ namespace Tensorflow.Keras.Layers
 
         protected override void build(TensorShape input_shape)
         {
-            var last_dim = input_shape.Dimensions.Last();
+            var last_dim = input_shape.dims.Last();
             var axes = new Dictionary<int, int>();
             axes[-1] = last_dim;
             input_spec = new InputSpec(min_ndim: 2, axes: axes);

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

@@ -30,7 +30,7 @@ namespace Tensorflow.Operations.Initializers
             if (dtype == TF_DataType.DtInvalid)
                 dtype = this.dtype;
 
-            return array_ops.ones(shape.Dimensions, dtype);
+            return array_ops.ones(shape.dims, dtype);
         }
 
         public object get_config()

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

@@ -132,9 +132,9 @@ namespace Tensorflow
             if(weights > 0)
             {
                 var weights_tensor = ops.convert_to_tensor(weights);
-                var labels_rank = labels.TensorShape.NDim;
+                var labels_rank = labels.TensorShape.ndim;
                 var weights_shape = weights_tensor.TensorShape;
-                var weights_rank = weights_shape.NDim;
+                var weights_rank = weights_shape.ndim;
 
                 if (labels_rank > -1 && weights_rank > -1)
                 {

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

@@ -35,18 +35,18 @@ namespace Tensorflow.Operations
             string name = null,
             string data_format = null)
         {
-            var num_total_dims = filter_shape.NDim;
+            var num_total_dims = filter_shape.ndim;
             var num_spatial_dims = num_total_dims - 2;
             int input_channels_dim;
             int[] spatial_dims;
             if (string.IsNullOrEmpty(data_format) || !data_format.StartsWith("NC"))
             {
-                input_channels_dim = input_shape.Dimensions[num_spatial_dims + 1];
+                input_channels_dim = input_shape.dims[num_spatial_dims + 1];
                 spatial_dims = Enumerable.Range(1, num_spatial_dims).ToArray();
             }
             else
             {
-                input_channels_dim = input_shape.Dimensions[1];
+                input_channels_dim = input_shape.dims[1];
                 spatial_dims = Enumerable.Range(2, num_spatial_dims).ToArray();
             }
 

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

@@ -36,7 +36,7 @@ namespace Tensorflow.Operations
         {
             this.padding = padding;
             this.name = name;
-            var conv_dims = input_shape.NDim - 2;
+            var conv_dims = input_shape.ndim - 2;
             if (conv_dims == 1)
             {
                 throw new NotImplementedException("_NonAtrousConvolution conv_dims 1");

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

@@ -33,7 +33,7 @@ namespace Tensorflow.Operations
         {
             var dilation_rate_tensor = ops.convert_to_tensor(dilation_rate, TF_DataType.TF_INT32, name: "dilation_rate");
             var rate_shape = dilation_rate_tensor.TensorShape;
-            var num_spatial_dims = rate_shape.Dimensions[0];
+            var num_spatial_dims = rate_shape.dims[0];
             int starting_spatial_dim = -1;
             if (!string.IsNullOrEmpty(data_format) && data_format.StartsWith("NC"))
                 starting_spatial_dim = 2;

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

@@ -19,6 +19,7 @@ using System.Collections.Generic;
 using System.Linq;
 using static Tensorflow.Python;
 using Tensorflow.Util;
+using NumSharp;
 
 namespace Tensorflow.Operations
 {
@@ -94,12 +95,12 @@ namespace Tensorflow.Operations
             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 dims = inputs_got_shape[0].Dimensions.Take(2).ToArray();
+            var dims = inputs_got_shape[0].dims.Take(2).ToArray();
             var (const_time_steps, const_batch_size) = (dims[0], dims[1]);
 
             foreach(var shape in inputs_got_shape)
             {
-                if (shape[2] == -1)
+                if (shape.dims[2] == -1)
                     throw new ValueError("Input size (depth of inputs) must be accessible via shape inference," +
                         " but saw value None.");
 
@@ -190,7 +191,7 @@ namespace Tensorflow.Operations
         public static Tensor _transpose_batch_time(Tensor x)
         {
             var x_static_shape = x.TensorShape;
-            if (x_static_shape.NDim == 1)
+            if (x_static_shape.ndim == 1)
                 return x;
 
             var x_rank = array_ops.rank(x);
@@ -201,9 +202,9 @@ namespace Tensorflow.Operations
             };
             var x_t = array_ops.transpose(x, array_ops.concat(con1, 0));
 
-            var dims = new int[] { x_static_shape.Dimensions[1], x_static_shape.Dimensions[0] }
+            var dims = new int[] { x_static_shape.dims[1], x_static_shape.dims[0] }
                 .ToList();
-            dims.AddRange(x_static_shape.Dimensions.Skip(2));
+            dims.AddRange(x_static_shape.dims.Skip(2));
             var shape = new TensorShape(dims.ToArray());
 
             x_t.SetShape(shape);
@@ -221,12 +222,12 @@ namespace Tensorflow.Operations
             foreach(var input_ in flat_input)
             {
                 var shape = input_.TensorShape;
-                if (shape.NDim < 0)
+                if (shape.ndim < 0)
                     continue;
-                if (shape.NDim < 2)
+                if (shape.ndim < 2)
                     throw new ValueError($"Expected input tensor {input_.name} to have rank at least 2");
 
-                var batch_size = shape.Dimensions[1];
+                var batch_size = shape.dims[1];
                 if (batch_size > -1)
                     throw new ValueError("_best_effort_input_batch_size batch_size > -1");
                     //return batch_size;

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

@@ -33,11 +33,11 @@ namespace Tensorflow.Operations
                 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_static = s_tensor_shape.NDim > -1 ?
-                s_tensor_shape.Dimensions :
+            var s_static = s_tensor_shape.ndim > -1 ?
+                s_tensor_shape.dims :
                 null;
             var s = s_tensor_shape.is_fully_defined() ?
-                constant_op.constant(s_tensor_shape.Dimensions, dtype: dtypes.int32) :
+                constant_op.constant(s_tensor_shape.dims, dtype: dtypes.int32) :
                 null;
 
             if (@static)

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

@@ -40,7 +40,7 @@ namespace Tensorflow
                 verify_shape: verify_shape,
                 allow_broadcast: false);
 
-        public static Tensor zeros(Shape shape, TF_DataType dtype = TF_DataType.TF_FLOAT, string name = null)
+        public static Tensor zeros(TensorShape shape, TF_DataType dtype = TF_DataType.TF_FLOAT, string name = null)
         {
             dtype = dtype.as_base_dtype();
             return with(ops.name_scope(name, "zeros", shape), scope =>
@@ -92,16 +92,16 @@ namespace Tensorflow
             return shape < 1000;
         }
 
-        private static Tensor _constant_if_small<T>(T value, Shape shape, TF_DataType dtype, string name)
+        private static Tensor _constant_if_small<T>(T value, TensorShape shape, TF_DataType dtype, string name)
         {
             Tensor tShape = null;
-            if (shape.Size < 1000)
+            if (shape.size < 1000)
             {
                 return constant_op.constant(value, shape: shape, dtype: dtype, name: name);
             }
             else
             {
-                tShape = constant_op._tensor_shape_tensor_conversion_function(shape.as_shape());
+                tShape = constant_op._tensor_shape_tensor_conversion_function(shape);
                 var c = constant_op.constant(0, dtype: dtype);
                 return gen_array_ops.fill(tShape, c, name: name);
             }
@@ -194,8 +194,8 @@ namespace Tensorflow
                 name = scope;
                 var input_tensor = ops.convert_to_tensor(input);
                 var input_shape = tensor_util.to_shape(input_tensor.shape);
-                if (optimize && input_shape.NDim > 0)
-                    return constant_op.constant(input_shape.NDim, dtype: tf.int32, name: name);
+                if (optimize && input_shape.ndim > 0)
+                    return constant_op.constant(input_shape.ndim, dtype: tf.int32, name: name);
                 else
                     return gen_array_ops.rank(input, name);
             });
@@ -372,7 +372,7 @@ namespace Tensorflow
                 {
                     if (input_shape.is_fully_defined())
                     {
-                        return constant_op.constant(input_shape.Size, dtype: out_type, name: name);
+                        return constant_op.constant(input_shape.size, dtype: out_type, name: name);
                     }
                 }
 

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

@@ -39,9 +39,9 @@ namespace Tensorflow
                 predictions = ops.convert_to_tensor(predictions);
                 labels = ops.convert_to_tensor(labels);
                 var predictions_shape = predictions.TensorShape;
-                var predictions_rank = predictions_shape.NDim;
+                var predictions_rank = predictions_shape.ndim;
                 var labels_shape = labels.TensorShape;
-                var labels_rank = labels_shape.NDim;
+                var labels_rank = labels_shape.ndim;
                 if(labels_rank > -1 && predictions_rank > -1)
                 {
                     // Use static rank.

src/TensorFlowNET.Core/Operations/nn_ops.cs

@@ -81,7 +81,7 @@ namespace Tensorflow
                 rate = ops.convert_to_tensor(rate, dtype: x.dtype, name: "rate");
                 // Do nothing if we know rate == 0
                 var val = tensor_util.constant_value(rate);
-                if (!(val is null) && val.Data<float>(0) == 0)
+                if (!(val is null) && val.Data<float>()[0] == 0)
                     return x;
 
                 noise_shape = _get_noise_shape(x, noise_shape);
@@ -185,7 +185,7 @@ namespace Tensorflow
                         logits.get_shape()[:-1].is_fully_defined());*/
 
                 // Check if no reshapes are required.
-                if(logits.TensorShape.NDim == 2)
+                if(logits.TensorShape.ndim == 2)
                 {
                     var (cost, _) = gen_nn_ops.sparse_softmax_cross_entropy_with_logits(
                         precise_logits, labels, name: name);
@@ -257,11 +257,11 @@ namespace Tensorflow
             // Set output shape if known.
             // if not context.executing_eagerly():
             var shape = logits.TensorShape;
-            if(shape != null && shape.NDim > 0)
+            if(shape != null && shape.ndim > 0)
             {
                 var product = 1;
                 var product_valid = true;
-                foreach(var d in shape.Dimensions.Take(shape.NDim - 1))
+                foreach(var d in shape.dims.Take(shape.ndim - 1))
                 {
                     if(d == -1)
                     {

src/TensorFlowNET.Core/Python.cs

@@ -15,6 +15,7 @@
 ******************************************************************************/
 
 using NumSharp;
+using NumSharp.Backends;
 using System;
 using System.Collections;
 using System.Collections.Generic;
@@ -48,7 +49,7 @@ namespace Tensorflow
                 case ICollection arr:
                     return arr.Count;
                 case NDArray ndArray:
-                    return ndArray.len;
+                    return ndArray.shape[0];
                 case IEnumerable enumerable:
                     return enumerable.OfType<object>().Count();
             }
@@ -141,9 +142,12 @@ namespace Tensorflow
         }
 
         public static IEnumerable<(T, T)> zip<T>(NDArray t1, NDArray t2)
+            where T : unmanaged
         {
-            for (int i = 0; i < t1.size; i++)
-                yield return (t1.Data<T>(i), t2.Data<T>(i));
+            var a = t1.AsIterator<T>();
+            var b = t2.AsIterator<T>();
+            while (a.HasNext())
+                yield return (a.MoveNext(), b.MoveNext());
         }
 
         public static IEnumerable<(T1, T2)> zip<T1, T2>(IList<T1> t1, IList<T2> t2)
@@ -152,10 +156,14 @@ namespace Tensorflow
                 yield return (t1[i], t2[i]);
         }
 
-        public static IEnumerable<(T1, T2)> zip<T1, T2>(NDArray t1, NDArray t2)
+        public static IEnumerable<(T1, T2)> zip<T1, T2>(NDArray t1, NDArray t2) 
+            where T1: unmanaged
+            where T2: unmanaged
         {
-            for (int i = 0; i < t1.size; i++)
-                yield return (t1.Data<T1>(i), t2.Data<T2>(i));
+            var a = t1.AsIterator<T1>();
+            var b = t2.AsIterator<T2>();
+            while(a.HasNext())
+                yield return (a.MoveNext(), b.MoveNext());
         }
 
         public static IEnumerable<(T1, T2)> zip<T1, T2>(IEnumerable<T1> e1, IEnumerable<T2> e2)

src/TensorFlowNET.Core/Sessions/_FetchHandler.cs

@@ -100,19 +100,19 @@ namespace Tensorflow
                         switch (value.dtype.Name)
                         {
                             case "Int32":
-                                full_values.Add(value.Data<int>(0));
+                                full_values.Add(value.Data<int>()[0]);
                                 break;
                             case "Int64":
-                                full_values.Add(value.Data<long>(0));
+                                full_values.Add(value.Data<long>()[0]);
                                 break;
                             case "Single":
-                                full_values.Add(value.Data<float>(0));
+                                full_values.Add(value.Data<float>()[0]);
                                 break;
                             case "Double":
-                                full_values.Add(value.Data<double>(0));
+                                full_values.Add(value.Data<double>()[0]);
                                 break;
                             case "String":
-                                full_values.Add(value.Data<string>(0));
+                                full_values.Add(value.Data<string>()[0]);
                                 break;
                         }
                     }

src/TensorFlowNET.Core/TensorFlowNET.Core.csproj

@@ -30,7 +30,7 @@ Docs: https://tensorflownet.readthedocs.io</Description>
 8. Upgrade NumSharp.
 9. Fix strided_slice_grad type convention error.
 10. Add AbsGrad.</PackageReleaseNotes>
-    <LangVersion>7.2</LangVersion>
+    <LangVersion>7.3</LangVersion>
     <FileVersion>0.10.3.0</FileVersion>
     <PackageLicenseFile>LICENSE</PackageLicenseFile>
     <PackageRequireLicenseAcceptance>true</PackageRequireLicenseAcceptance>
@@ -63,11 +63,14 @@ Docs: https://tensorflownet.readthedocs.io</Description>
 
   <ItemGroup>
     <PackageReference Include="Google.Protobuf" Version="3.9.0" />
-    <PackageReference Include="NumSharp" Version="0.10.6" />
   </ItemGroup>
 
   <ItemGroup>
     <Folder Include="Distribute\" />
     <Folder Include="Keras\Initializers\" />
   </ItemGroup>
+
+  <ItemGroup>
+    <ProjectReference Include="..\..\..\NumSharp\src\NumSharp.Core\NumSharp.Core.csproj" />
+  </ItemGroup>
 </Project>

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

@@ -513,7 +513,7 @@ namespace Tensorflow
                     Marshal.Copy(nd1.Data<byte>(), 0, dotHandle, nd.size);
                     break;
                 case "String":
-                    return new Tensor(UTF8Encoding.UTF8.GetBytes(nd.Data<string>(0)), TF_DataType.TF_STRING);
+                    return new Tensor(UTF8Encoding.UTF8.GetBytes(nd.Data<string>()[0]), TF_DataType.TF_STRING);
                 default:
                     throw new NotImplementedException($"Marshal.Copy failed for {nd.dtype.Name}.");
             }

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

@@ -77,12 +77,12 @@ namespace Tensorflow
                 throw new ArgumentNullException(nameof(tensor));
             }
 
-            if (tensor.TensorShape.NDim != 0)
+            if (tensor.TensorShape.ndim != 0)
             {
                 throw new ArgumentException("Tensor must have 0 dimensions in order to convert to scalar");
             }
 
-            if (tensor.TensorShape.Size != 1)
+            if (tensor.TensorShape.size != 1)
             {
                 throw new ArgumentException("Tensor must have size 1 in order to convert to scalar");
             }

src/TensorFlowNET.Core/Tensors/Tensor.cs

@@ -105,9 +105,9 @@ namespace Tensorflow
 
         public TensorShape TensorShape => tensor_util.to_shape(shape);
 
-        public void SetShape(Shape shape)
+        public void SetShape(TensorShape shape)
         {
-            this.shape = shape.Dimensions;
+            this.shape = shape.dims;
         }
 
         public void set_shape(Tensor shape)

src/TensorFlowNET.Core/Tensors/TensorShape.cs

@@ -7,27 +7,30 @@ namespace Tensorflow
     /// <summary>
     /// Represents the shape of a `Tensor`.
     /// </summary>
-    public class TensorShape : Shape
+    public class TensorShape
     {
-        public int[] dims => Dimensions;
+        private Shape shape;
+        public int[] dims => shape.Dimensions;
+        public int ndim => shape.NDim;
+        public int size => shape.Size;
 
         public TensorShape(TensorShapeProto proto)
         {
             if (proto.UnknownRank) return;
 
-            Reshape(proto.Dim.Select(x => (int)x.Size).ToArray());
+            shape.reshape(proto.Dim.Select(x => (int)x.Size).ToArray());
         }
 
-        public TensorShape(params int[] dims) : base(dims)
+        public TensorShape(params int[] dims)
         {
-
+            shape = new Shape(dims);
         }
 
         public TensorShape this[Slice slice]
         {
             get
             {
-                return new TensorShape(Dimensions.Skip(slice.Start.Value)
+                return new TensorShape(dims.Skip(slice.Start.Value)
                     .Take(slice.Length.Value)
                     .ToArray());
             }
@@ -39,7 +42,7 @@ namespace Tensorflow
         /// <returns></returns>
         public bool is_fully_defined()
         {
-            return Dimensions != null && Dimensions.Count(x => x < 1) == 0;
+            return dims != null && dims.Count(x => x < 1) == 0;
         }
 
         public bool is_compatible_with(TensorShape shape2)
@@ -49,7 +52,7 @@ namespace Tensorflow
 
         public TensorShape with_rank_at_least(int rank)
         {
-            if (rank != this.NDim)
+            if (rank != ndim)
                 throw new ValueError($"Shape {this} must have rank at least {rank}");
             else
                 return this;
@@ -64,13 +67,16 @@ namespace Tensorflow
         {
             var other = new TensorShape(other_);
 
-            if (NDim < 0 || other.NDim < 0)
+            if (ndim < 0 || other.ndim < 0)
                 return new TensorShape();
             else
-                return new TensorShape(NDim + other.NDim);
+                return new TensorShape(ndim + other.ndim);
         }
 
+        public static implicit operator TensorShape(Shape shape) => new TensorShape(shape.Dimensions);
+        public static implicit operator Shape(TensorShape shape) => new Shape(shape.dims);
         public static implicit operator TensorShape(int[] dims) => new TensorShape(dims);
+        public static implicit operator int[](TensorShape shape) => shape.dims;
         public static implicit operator TensorShape((int, int) dims) => new TensorShape(dims.Item1, dims.Item2);
         public static implicit operator TensorShape((int, int, int) dims) => new TensorShape(dims.Item1, dims.Item2, dims.Item3);
         public static implicit operator TensorShape((int, int, int, int) dims) => new TensorShape(dims.Item1, dims.Item2, dims.Item3, dims.Item4);

src/TensorFlowNET.Core/Tensors/constant_op.cs

@@ -81,7 +81,7 @@ namespace Tensorflow
         /// <returns></returns>
         public static Tensor _tensor_shape_tensor_conversion_function(TensorShape s, TF_DataType dtype = TF_DataType.DtInvalid, string name = null, bool as_ref = false)
         {
-            var s_list = s.Dimensions;
+            var s_list = s.dims;
             var int64_value = 0;
             foreach(var dim in s_list)
             {

src/TensorFlowNET.Core/Tensors/tensor_util.cs

@@ -247,7 +247,7 @@ namespace Tensorflow
             }
             else
             {
-                shape_size = new TensorShape(shape).Size;
+                shape_size = new TensorShape(shape).size;
                 is_same_size = shape_size == nparray.size;
             }
 
@@ -387,10 +387,10 @@ namespace Tensorflow
         {
             TensorShapeProto shape = new TensorShapeProto();
 
-            for (int i = 0; i < tshape.NDim; i++)
+            for (int i = 0; i < tshape.ndim; i++)
             {
                 var dim = new TensorShapeProto.Types.Dim();
-                dim.Size = tshape.Dimensions[i];
+                dim.Size = tshape.dims[i];
                 //dim.Name = $"dim_{i}";
 
                 shape.Dim.Add(dim);

src/TensorFlowNET.Core/Tensors/tf.constant.cs

@@ -42,10 +42,10 @@ namespace Tensorflow
                 verify_shape: false,
                 allow_broadcast: false);
 
-        public static Tensor zeros(Shape shape, TF_DataType dtype = TF_DataType.TF_FLOAT, string name = null) 
+        public static Tensor zeros(TensorShape shape, TF_DataType dtype = TF_DataType.TF_FLOAT, string name = null) 
             => array_ops.zeros(shape, dtype, name);
 
-        public static Tensor ones(Shape shape, TF_DataType dtype = TF_DataType.TF_FLOAT, string name = null) 
+        public static Tensor ones(TensorShape shape, TF_DataType dtype = TF_DataType.TF_FLOAT, string name = null) 
             => array_ops.ones(shape, dtype, name);
 
         public static Tensor size(Tensor input,

src/TensorFlowNET.Core/Variables/_VariableStore.cs

@@ -67,7 +67,7 @@ namespace Tensorflow
             VariableSynchronization synchronization = VariableSynchronization.Auto,
             VariableAggregation aggregation = VariableAggregation.None)
         {
-            bool is_scalar = !(shape is null) && shape.NDim == 0;
+            bool is_scalar = !(shape is null) && shape.ndim == 0;
 
             if (initializer is IInitializer init)
             {

test/TensorFlowNET.Examples/BasicModels/KMeansClustering.cs

@@ -130,7 +130,7 @@ namespace TensorFlowNET.Examples
             var counts = np.zeros((k, num_classes), np.float32);
 
             sw.Start();
-            foreach (var i in range(idx.Length))
+            foreach (var i in range(idx.Count))
             {
                 var x = mnist.train.labels[i];
                 counts[idx[i]] += x;

test/TensorFlowNET.Examples/BasicModels/NaiveBayesClassifier.cs

@@ -42,10 +42,10 @@ namespace TensorFlowNET.Examples
             fit(X, y);
 
             // Create a regular grid and classify each point 
-            float x_min = X.amin(0).Data<float>(0) - 0.5f;
-            float y_min = X.amin(0).Data<float>(1) - 0.5f;
-            float x_max = X.amax(0).Data<float>(0) + 0.5f;
-            float y_max = X.amax(0).Data<float>(1) + 0.5f;
+            float x_min = X.amin(0).Data<float>()[0] - 0.5f;
+            float y_min = X.amin(0).Data<float>()[1] - 0.5f;
+            float x_max = X.amax(0).Data<float>()[1] + 0.5f;
+            float y_max = X.amax(0).Data<float>()[1] + 0.5f;
 
             var (xx, yy) = np.meshgrid(np.linspace(x_min, x_max, 30), np.linspace(y_min, y_max, 30));
             with(tf.Session(), sess =>

test/TensorFlowNET.Examples/ImageProcessing/DigitRecognitionCNN.cs

@@ -243,7 +243,7 @@ namespace TensorFlowNET.Examples.ImageProcess
             return with(tf.variable_scope("Flatten_layer"), delegate
             {
                 var layer_shape = layer.TensorShape;
-                var num_features = layer_shape[new Slice(1, 4)].Size;
+                var num_features = layer_shape[new Slice(1, 4)].size;
                 var layer_flat = tf.reshape(layer, new[] { -1, num_features });
 
                 return layer_flat;

test/TensorFlowNET.Examples/ImageProcessing/RetrainImageClassifier.cs

@@ -179,12 +179,12 @@ namespace TensorFlowNET.Examples.ImageProcess
         private (Operation, Tensor, Tensor, Tensor, Tensor) add_final_retrain_ops(int class_count, string final_tensor_name,
             Tensor bottleneck_tensor, bool quantize_layer, bool is_training)
         {
-            var (batch_size, bottleneck_tensor_size) = (bottleneck_tensor.TensorShape.Dimensions[0], bottleneck_tensor.TensorShape.Dimensions[1]);
+            var (batch_size, bottleneck_tensor_size) = (bottleneck_tensor.TensorShape.dims[0], bottleneck_tensor.TensorShape.dims[1]);
             with(tf.name_scope("input"), scope =>
             {
                 bottleneck_input = tf.placeholder_with_default(
                     bottleneck_tensor,
-                    shape: bottleneck_tensor.TensorShape.Dimensions,
+                    shape: bottleneck_tensor.TensorShape.dims,
                     name: "BottleneckInputPlaceholder");
 
                 ground_truth_input = tf.placeholder(tf.int64, new TensorShape(batch_size), name: "GroundTruthInput");
@@ -206,7 +206,7 @@ namespace TensorFlowNET.Examples.ImageProcess
                 RefVariable layer_biases = null;
                 with(tf.name_scope("biases"), delegate
                 {
-                    layer_biases = tf.Variable(tf.zeros(class_count), name: "final_biases");
+                    layer_biases = tf.Variable(tf.zeros(new TensorShape(class_count)), name: "final_biases");
                     variable_summaries(layer_biases);
                 });
 
@@ -662,7 +662,7 @@ namespace TensorFlowNET.Examples.ImageProcess
                 var train_summary = results[0];
 
                 // TODO
-                train_writer.add_summary(train_summary, i);
+                // train_writer.add_summary(train_summary, i);
 
                 // Every so often, print out how well the graph is training.
                 bool is_last_step = (i + 1 == how_many_training_steps);
@@ -687,9 +687,9 @@ namespace TensorFlowNET.Examples.ImageProcess
                         new FeedItem(bottleneck_input, validation_bottlenecks),
                         new FeedItem(ground_truth_input, validation_ground_truth));
 
-                    (string validation_summary, float validation_accuracy) = (results[0], results[1]);
-
-                    validation_writer.add_summary(validation_summary, i);
+                    //(string validation_summary, float validation_accuracy) = (results[0], results[1]);
+                    float validation_accuracy = results[1];
+                    // validation_writer.add_summary(validation_summary, i);
                     print($"{DateTime.Now}: Step {i + 1}: Validation accuracy = {validation_accuracy * 100}% (N={len(validation_bottlenecks)}) {sw.ElapsedMilliseconds}ms");
                     sw.Restart();
                 }

test/TensorFlowNET.Examples/TextProcessing/BinaryTextClassification.cs

@@ -86,11 +86,11 @@ namespace TensorFlowNET.Examples
             // convert x_train
             train_data = new NDArray(np.int32, (x_train.size, 256));
             for (int i = 0; i < x_train.size; i++)
-                train_data[i] = x_train[i].Data<string>(0).Split(',').Select(x => int.Parse(x)).ToArray();
+                train_data[i] = x_train[i].Data<string>()[1].Split(',').Select(x => int.Parse(x)).ToArray();
 
             test_data = new NDArray(np.int32, (x_test.size, 256));
             for (int i = 0; i < x_test.size; i++)
-                test_data[i] = x_test[i].Data<string>(0).Split(',').Select(x => int.Parse(x)).ToArray();
+                test_data[i] = x_test[i].Data<string>()[1].Split(',').Select(x => int.Parse(x)).ToArray();
 
             train_labels = y_train;
             test_labels = y_test;

test/TensorFlowNET.Examples/Utility/MNIST.cs

@@ -111,10 +111,10 @@ namespace TensorFlowNET.Examples.Utility
             var num_labels = labels_dense.shape[0];
             var index_offset = np.arange(num_labels) * num_classes;
             var labels_one_hot = np.zeros(num_labels, num_classes);
-
-            for(int row = 0; row < num_labels; row++)
+            var bytes = labels_dense.Data<byte>();
+            for (int row = 0; row < num_labels; row++)
             {
-                var col = labels_dense.Data<byte>(row);
+                var col = bytes[row];
                 labels_one_hot.SetData(1.0, row, col);
             }
 
@@ -125,7 +125,7 @@ namespace TensorFlowNET.Examples.Utility
         {
             var buffer = new byte[sizeof(uint)];
             var count = bytestream.Read(buffer, 0, 4);
-            return np.frombuffer(buffer, ">u4").Data<uint>(0);
+            return np.frombuffer(buffer, ">u4").Data<uint>()[0];
         }
     }
 }