add data type when determine which property should be taken out. #115

Author: Oceania2018

src/TensorFlowNET.Core/Eager/Tape.cs

@@ -0,0 +1,27 @@
+using System;
+using System.Collections.Generic;
+using System.Text;
+
+namespace Tensorflow.Eager
+{
+    public class Tape
+    {
+        public static bool IsDtypeTrainable(DataType dtype)
+        {
+            switch (dtype)
+            {
+                case DataType.DtHalf:
+                case DataType.DtBfloat16:
+                case DataType.DtFloat:
+                case DataType.DtDouble:
+                case DataType.DtComplex64:
+                case DataType.DtComplex128:
+                case DataType.DtResource:
+                case DataType.DtVariant:
+                    return true;
+                default:
+                    return false;
+            }
+        }
+    }
+}

src/TensorFlowNET.Core/Eager/pywrap_tfe_src.cs

@@ -12,7 +12,19 @@ public class pywrap_tfe_src
     {
         public static void RecordGradient(string op_name, Tensor[] inputs, Dictionary<string, object> attrs, Tensor[] results, string name = "")
         {
-            
+            var input_ids = inputs.Select(x => x.Id).ToArray();
+            var input_dtypes = inputs.Select(x => x.dtype).ToArray();
+
+            bool should_record = false;
+            foreach (var input_dtype in input_dtypes)
+            {
+                if (Tape.IsDtypeTrainable(input_dtype.as_datatype_enum()))
+                {
+                    should_record = true;
+                    break;
+                }
+            }
+            if (!should_record) return;
         }
     }
 }

src/TensorFlowNET.Core/Operations/Operation.cs

@@ -1,5 +1,6 @@
 using System;
 using System.Collections.Generic;
+using System.Linq;
 using System.Runtime.InteropServices;
 using System.Text;
 
@@ -126,11 +127,11 @@ public Operation(NodeDef node_def, Graph g, List<Tensor> inputs = null, TF_DataT
             Graph._add_op(this);
         }
 
-        public object get_attr(string name)
+        public object get_attr<T>(string name)
         {
             AttrValue x = null;
 
-            var fields = new string[] { "s", "i", "f", "b", "type", "shape", "tensor", "func" };
+            var fields = new string[] { "s", "i", "f", "b", "Type", "Shape", "Tensor", "func" };
 
             using (var buf = new Buffer())
             {
@@ -141,12 +142,21 @@ public object get_attr(string name)
 
             switch (name)
             {
+                case "T":
                 case "dtype":
                     return x.Type;
                 case "shape":
                     return x.Shape;
                 default:
-                    throw new NotImplementedException($"{name}");
+                    switch (typeof(T).Name)
+                    {
+                        case "Boolean":
+                            return x.B;
+                        case "String":
+                            return x.S;
+                        default:
+                            throw new NotImplementedException($"Unsupported field type in {x.ToString()}");
+                    }
             }
         }
 

src/TensorFlowNET.Core/Operations/gen_array_ops.cs

@@ -21,12 +21,13 @@ public static Tensor placeholder(TF_DataType dtype, TensorShape shape = null, st
             var _op = _op_def_lib._apply_op_helper("Placeholder", keywords: keywords);
             var _result = _op.outputs;
             var _inputs_flat = _op.inputs;
-            var _attrs = new Dictionary<string, object>();
 
-            _attrs["dtype"] = _op.get_attr("dtype");
-            _attrs["shape"] = _op.get_attr("shape");
+            var _attrs = new Dictionary<string, object>();
+            _attrs["dtype"] = _op.get_attr<DataType>("dtype");
+            _attrs["shape"] = _op.get_attr<int[]>("shape");
 
             _execute.record_gradient("Placeholder", _inputs_flat, _attrs, _result, name);
+
             return new Tensor(_op, 0, dtype);
         }
 

src/TensorFlowNET.Core/Tensors/Tensor.cs

@@ -16,6 +16,9 @@ public partial class Tensor : IDisposable
     {
         private readonly IntPtr _handle;
 
+        private int _id;
+        public int Id => _id;
+
         public Graph Graph => op.Graph;
         public Operation op { get; }
 

src/TensorFlowNET.Core/Variables/gen_state_ops.py.cs

@@ -2,12 +2,14 @@
 using System;
 using System.Collections.Generic;
 using System.Text;
+using Tensorflow.Eager;
 
 namespace Tensorflow
 {
     public class gen_state_ops
     {
         public static OpDefLibrary _op_def_lib = new OpDefLibrary();
+        public static Execute _execute = new Execute();
 
         /// <summary>
         /// Holds state in the form of a tensor that persists across steps.
@@ -32,6 +34,14 @@ public static Tensor variable_v2(long[] shape, TF_DataType dtype, string name =
             var _result = _op.outputs;
             var _inputs_flat = _op.inputs;
 
+            var _attrs = new Dictionary<string, object>();
+            _attrs["dtype"] = _op.get_attr<DataType>("dtype");
+            _attrs["shape"] = _op.get_attr<int[]>("shape");
+            _attrs["container"] = _op.get_attr<string>("container");
+            _attrs["shared_name"] = _op.get_attr<string>("shared_name");
+
+            _execute.record_gradient("Placeholder", _inputs_flat, _attrs, _result, name);
+
             return new Tensor(_op, 0, dtype);
         }
 
@@ -56,9 +66,17 @@ public static Tensor assign(Tensor tensor, Tensor value,
 
             var _op = _op_def_lib._apply_op_helper("Assign", name: name, keywords: keywords);
 
-            var _result = _op.outputs[0];
+            var _result = _op.outputs;
             var _inputs_flat = _op.inputs;
-            return _result;
+
+            var _attrs = new Dictionary<string, object>();
+            _attrs["T"] = _op.get_attr<DataType>("T");
+            _attrs["validate_shape"] = _op.get_attr<bool>("validate_shape");
+            _attrs["use_locking"] = _op.get_attr<bool>("use_locking");
+
+            _execute.record_gradient("Placeholder", _inputs_flat, _attrs, _result, name);
+
+            return _result[0];
         }
     }
 }