fixed #175

src/TensorFlowNET.Core/Gradients/math_grad.py.cs

@@ -47,10 +47,14 @@ namespace Tensorflow
             x = math_ops.conj(x);
             y = math_ops.conj(y);
 
-            var r1 = math_ops.reduce_sum(gen_math_ops.mul(grad, y), rx);
-            var r2 = math_ops.reduce_sum(gen_math_ops.mul(x, grad), ry);
-
-            return (gen_array_ops.reshape(r1, sx), gen_array_ops.reshape(r2, sy));
+            var mul1 = gen_math_ops.mul(grad, y);
+            var mul2 = gen_math_ops.mul(x, grad);
+            var reduce_sum1 = math_ops.reduce_sum(mul1, rx);
+            var reduce_sum2 = math_ops.reduce_sum(mul2, ry);
+            var reshape1 = gen_array_ops.reshape(reduce_sum1, sx);
+            var reshape2 = gen_array_ops.reshape(reduce_sum2, sy);
+
+            return (reshape1, reshape2);
         }
 
         public static (Tensor, Tensor) _SubGrad(Operation op, Tensor grad)
@@ -129,9 +133,12 @@ namespace Tensorflow
             var (rx, ry) = gen_array_ops.broadcast_gradient_args(sx, sy);
             x = math_ops.conj(x);
             y = math_ops.conj(y);
-            y = math_ops.conj(z);
-            var gx = gen_array_ops.reshape(math_ops.reduce_sum(grad * y * gen_math_ops.pow(x, y - 1.0), rx), sx);
-            Tensor log_x = null;
+            z = math_ops.conj(z);
+            var pow = gen_math_ops.pow(x, y - 1.0f);
+            var mul = grad * y * pow;
+            var reduce_sum = math_ops.reduce_sum(mul, rx);
+            var gx = gen_array_ops.reshape(reduce_sum, sx);
+
             // Avoid false singularity at x = 0
             Tensor mask = null;
             if (x.dtype.is_complex())
@@ -142,8 +149,10 @@ namespace Tensorflow
             var safe_x = array_ops.where(mask, x, ones);
             var x1 = gen_array_ops.log(safe_x);
             var y1 = array_ops.zeros_like(x);
-            log_x = array_ops.where(mask, x1, y1);
-            var gy = gen_array_ops.reshape(math_ops.reduce_sum(grad * z * log_x, ry), sy);
+            var log_x = array_ops.where(mask, x1, y1);
+            var mul1 = grad * z * log_x;
+            var reduce_sum1 = math_ops.reduce_sum(mul1, ry);
+            var gy = gen_array_ops.reshape(reduce_sum1, sy);
 
             return (gx, gy);
         }

src/TensorFlowNET.Core/Graphs/Graph.cs

@@ -196,11 +196,11 @@ namespace Tensorflow
 
             _create_op_helper(op, true);
 
-            Console.Write($"create_op: {op_type} '{node_def.Name}'");
+            /*Console.Write($"create_op: {op_type} '{node_def.Name}'");
             Console.Write($", inputs: {(inputs.Length == 0 ? "empty" : String.Join(", ", inputs.Select(x => x.name)))}");
             Console.Write($", control_inputs: {(control_inputs.Length == 0 ? "empty" : String.Join(", ", control_inputs.Select(x => x.name)))}");
             Console.Write($", outputs: {(op.outputs.Length == 0 ? "empty" : String.Join(", ", op.outputs.Select(x => x.name)))}");
-            Console.WriteLine();
+            Console.WriteLine();*/
 
             return op;
         }

src/TensorFlowNET.Core/Operations/Operation.Output.cs

@@ -1,5 +1,4 @@
-using Newtonsoft.Json;
-using System;
+using System;
 using System.Collections.Generic;
 using System.Linq;
 using System.Runtime.InteropServices;
@@ -15,7 +14,7 @@ namespace Tensorflow
 
         private Tensor[] _outputs;
         public Tensor[] outputs => _outputs;
-        [JsonIgnore]
+        //[JsonIgnore]
         public Tensor output => _outputs.FirstOrDefault();
 
         public int NumControlOutputs => c_api.TF_OperationNumControlOutputs(_handle);

src/TensorFlowNET.Core/Operations/Operation.cs

@@ -1,5 +1,4 @@
 using Google.Protobuf.Collections;
-using Newtonsoft.Json;
 using System;
 using System.Collections.Generic;
 using System.Linq;
@@ -13,15 +12,15 @@ namespace Tensorflow
         private readonly IntPtr _handle; // _c_op in python
 
         private Graph _graph;
-        [JsonIgnore]
+        //[JsonIgnore]
         public Graph graph => _graph;
-        [JsonIgnore]
+        //[JsonIgnore]
         public int _id => _id_value;
-        [JsonIgnore]
+        //[JsonIgnore]
         public int _id_value;
 
         public string type => OpType;
-        [JsonIgnore]
+        //[JsonIgnore]
         public Operation op => this;
         public TF_DataType dtype => TF_DataType.DtInvalid;
         private Status status = new Status();

src/TensorFlowNET.Core/TensorFlowNET.Core.csproj

@@ -52,10 +52,4 @@ Upgraded to TensorFlow 1.13 RC2.
     <Content CopyToOutputDirectory="PreserveNewest" Include="./runtimes/win-x64/native/tensorflow.dll" Link="tensorflow.dll" Pack="true" PackagePath="runtimes/win-x64/native/tensorflow.dll" />
   </ItemGroup>
 
-  <ItemGroup>
-    <Reference Include="Newtonsoft.Json">
-      <HintPath>C:\Program Files\dotnet\sdk\NuGetFallbackFolder\newtonsoft.json\9.0.1\lib\netstandard1.0\Newtonsoft.Json.dll</HintPath>
-    </Reference>
-  </ItemGroup>
-
 </Project>

src/TensorFlowNET.Core/Tensors/Tensor.cs

@@ -1,5 +1,4 @@
-using Newtonsoft.Json;
-using NumSharp.Core;
+using NumSharp.Core;
 using System;
 using System.Collections.Generic;
 using System.Linq;
@@ -18,13 +17,13 @@ namespace Tensorflow
         private readonly IntPtr _handle;
 
         private int _id;
-        [JsonIgnore]
+        //[JsonIgnore]
         public int Id => _id;
-        [JsonIgnore]
+        //[JsonIgnore]
         public Graph graph => op?.graph;
-        [JsonIgnore]
+        //[JsonIgnore]
         public Operation op { get; }
-        [JsonIgnore]
+        //[JsonIgnore]
         public Tensor[] outputs => op.outputs;
 
         /// <summary>
@@ -104,7 +103,7 @@ namespace Tensorflow
 
         public int NDims => rank;
 
-        [JsonIgnore]
+        //[JsonIgnore]
         public Operation[] Consumers => consumers();
 
         public string Device => op.Device;

src/TensorFlowNET.Core/ops.py.cs

@@ -351,7 +351,7 @@ namespace Tensorflow
 
             return (oper, out_grads) =>
             {
-                Console.WriteLine($"get_gradient_function: {oper.type} '{oper.name}'");
+                // Console.WriteLine($"get_gradient_function: {oper.type} '{oper.name}'");
 
                 switch (oper.type)
                 {

test/TensorFlowNET.Examples/LinearRegression.cs

@@ -1,5 +1,4 @@
-using Newtonsoft.Json;
-using NumSharp.Core;
+using NumSharp.Core;
 using System;
 using System.Collections.Generic;
 using System.Text;
@@ -13,17 +12,15 @@ namespace TensorFlowNET.Examples
     /// </summary>
     public class LinearRegression : Python, IExample
     {
-        private NumPyRandom rng = np.random;
+        NumPyRandom rng = np.random;
+
+        // Parameters
+        float learning_rate = 0.01f;
+        int training_epochs = 1000;
+        int display_step = 50;
 
         public void Run()
         {
-            var graph = tf.Graph().as_default();
-
-            // Parameters
-            float learning_rate = 0.01f;
-            int training_epochs = 1000;
-            int display_step = 10;
-
             // Training Data
             var train_X = np.array(3.3f, 4.4f, 5.5f, 6.71f, 6.93f, 4.168f, 9.779f, 6.182f, 7.59f, 2.167f,
                          7.042f, 10.791f, 5.313f, 7.997f, 5.654f, 9.27f, 3.1f);
@@ -31,46 +28,28 @@ namespace TensorFlowNET.Examples
                          2.827f, 3.465f, 1.65f, 2.904f, 2.42f, 2.94f, 1.3f);
             var n_samples = train_X.shape[0];
 
+            var graph = tf.Graph().as_default();
+
             // tf Graph Input
             var X = tf.placeholder(tf.float32);
             var Y = tf.placeholder(tf.float32);
 
             // Set model weights 
-            //var rnd1 = rng.randn<float>();
-            //var rnd2 = rng.randn<float>();
+            // We can set a fixed init value in order to debug
+            // var rnd1 = rng.randn<float>();
+            // var rnd2 = rng.randn<float>();
             var W = tf.Variable(-0.06f, name: "weight");
             var b = tf.Variable(-0.73f, name: "bias");
 
-            var mul = tf.multiply(X, W);
-            var pred = tf.add(mul, b);
+            // Construct a linear model
+            var pred = tf.add(tf.multiply(X, W), b);
 
             // Mean squared error
-            var sub = pred - Y;
-            var pow = tf.pow(sub, 2.0f);
-
-            var reduce = tf.reduce_sum(pow);
-            var cost = reduce / (2.0f * n_samples);
+            var cost = tf.reduce_sum(tf.pow(pred - Y, 2.0f)) / (2.0f * n_samples);
 
             // radient descent
             // Note, minimize() knows to modify W and b because Variable objects are trainable=True by default
-            var grad = tf.train.GradientDescentOptimizer(learning_rate);
-            var optimizer = grad.minimize(cost);
-
-            //tf.train.export_meta_graph(filename: "linear_regression.meta.bin");
-            // import meta
-            // var new_saver = tf.train.import_meta_graph("linear_regression.meta.bin");
-            var text = JsonConvert.SerializeObject(graph, new JsonSerializerSettings
-            {
-                Formatting = Formatting.Indented
-            });
-
-            /*var cost = graph.OperationByName("truediv").output;
-            var pred = graph.OperationByName("Add").output;
-            var optimizer = graph.OperationByName("GradientDescent");
-            var X = graph.OperationByName("Placeholder").output;
-            var Y = graph.OperationByName("Placeholder_1").output;
-            var W = graph.OperationByName("weight").output;
-            var b = graph.OperationByName("bias").output;*/
+            var optimizer = tf.train.GradientDescentOptimizer(learning_rate).minimize(cost);
 
             // Initialize the variables (i.e. assign their default value)
             var init = tf.global_variables_initializer();
@@ -89,22 +68,33 @@ namespace TensorFlowNET.Examples
                         sess.run(optimizer, 
                             new FeedItem(X, x),
                             new FeedItem(Y, y));
-                        var rW = sess.run(W);
                     }
 
                     // Display logs per epoch step
-                    /*if ((epoch + 1) % display_step == 0)
+                    if ((epoch + 1) % display_step == 0)
                     {
                         var c = sess.run(cost, 
                             new FeedItem(X, train_X),
                             new FeedItem(Y, train_Y));
-                        var rW = sess.run(W);
-                        Console.WriteLine($"Epoch: {epoch + 1} cost={c} " +
-                                    $"W={rW} b={sess.run(b)}");
-                    }*/
+                        Console.WriteLine($"Epoch: {epoch + 1} cost={c} " + $"W={sess.run(W)} b={sess.run(b)}");
+                    }
                 }
 
                 Console.WriteLine("Optimization Finished!");
+                var training_cost = sess.run(cost,
+                    new FeedItem(X, train_X),
+                    new FeedItem(Y, train_Y));
+                Console.WriteLine($"Training cost={training_cost} W={sess.run(W)} b={sess.run(b)}");
+
+                // Testing example
+                var test_X = np.array(6.83f, 4.668f, 8.9f, 7.91f, 5.7f, 8.7f, 3.1f, 2.1f);
+                var test_Y = np.array(1.84f, 2.273f, 3.2f, 2.831f, 2.92f, 3.24f, 1.35f, 1.03f);
+                Console.WriteLine("Testing... (Mean square loss Comparison)");
+                var testing_cost = sess.run(tf.reduce_sum(tf.pow(pred - Y, 2.0f)) / (2.0f * test_X.shape[0]),
+                    new FeedItem(X, test_X), 
+                    new FeedItem(Y, test_Y));
+                Console.WriteLine($"Testing cost={testing_cost}");
+                Console.WriteLine($"Absolute mean square loss difference: {Math.Abs((float)training_cost - (float)testing_cost)}");
             });
         }
     }

test/TensorFlowNET.Examples/TensorFlowNET.Examples.csproj

@@ -6,7 +6,6 @@
   </PropertyGroup>
 
   <ItemGroup>
-    <PackageReference Include="Newtonsoft.Json" Version="12.0.1" />
     <PackageReference Include="NumSharp" Version="0.7.3" />
     <PackageReference Include="TensorFlow.NET" Version="0.3.0" />
   </ItemGroup>

test/TensorFlowNET.UnitTest/TrainSaverTest.cs

@@ -23,6 +23,23 @@ namespace TensorFlowNET.UnitTest
             {
                 var new_saver = tf.train.import_meta_graph("C:/tmp/my-model.meta");
             });
+
+            //tf.train.export_meta_graph(filename: "linear_regression.meta.bin");
+            // import meta
+            /*tf.train.import_meta_graph("linear_regression.meta.bin");
+
+            var cost = graph.OperationByName("truediv").output;
+            var pred = graph.OperationByName("Add").output;
+            var optimizer = graph.OperationByName("GradientDescent");
+            var X = graph.OperationByName("Placeholder").output;
+            var Y = graph.OperationByName("Placeholder_1").output;
+            var W = graph.OperationByName("weight").output;
+            var b = graph.OperationByName("bias").output;*/
+
+            /*var text = JsonConvert.SerializeObject(graph, new JsonSerializerSettings
+            {
+                Formatting = Formatting.Indented
+            });*/
         }
 
         public void ImportSavedModel()