node_def property in Operation #134

src/TensorFlowNET.Core/Operations/Operation.cs

@@ -36,7 +36,7 @@ namespace Tensorflow
             var handle = Marshal.AllocHGlobal(size);
             int num = c_api.TF_OperationOutputConsumers(new TF_Output(_handle, index), handle, max_consumers);
             var consumers = new TF_Input[num];
-            for(int i = 0; i < num; i++)
+            for (int i = 0; i < num; i++)
             {
                 consumers[i] = Marshal.PtrToStructure<TF_Input>(handle + i * size);
             }
@@ -50,7 +50,7 @@ namespace Tensorflow
         {
             var control_inputs = new Operation[NumControlInputs];
 
-            if(NumControlInputs > 0)
+            if (NumControlInputs > 0)
             {
                 IntPtr control_input_handle = Marshal.AllocHGlobal(Marshal.SizeOf<IntPtr>() * NumControlInputs);
                 c_api.TF_OperationGetControlInputs(_handle, control_input_handle, NumControlInputs);
@@ -70,7 +70,7 @@ namespace Tensorflow
         {
             var control_outputs = new Operation[NumControlOutputs];
 
-            if(NumControlOutputs > 0)
+            if (NumControlOutputs > 0)
             {
                 IntPtr control_output_handle = Marshal.AllocHGlobal(Marshal.SizeOf<IntPtr>() * NumControlOutputs);
                 c_api.TF_OperationGetControlOutputs(_handle, control_output_handle, NumControlInputs);
@@ -89,7 +89,7 @@ namespace Tensorflow
         {
             get
             {
-                if(_outputs == null)
+                if (_outputs == null)
                 {
                     _outputs = new Tensor[NumOutputs];
 
@@ -106,7 +106,7 @@ namespace Tensorflow
         {
             get
             {
-                if(_inputs == null)
+                if (_inputs == null)
                 {
                     var retval = new Tensor[NumInputs];
 
@@ -124,6 +124,18 @@ namespace Tensorflow
             }
         }
 
+        private NodeDef _node_def;
+        public NodeDef node_def
+        {
+            get
+            {
+                if(_node_def == null)
+                    _node_def = GetNodeDef();
+
+                return _node_def;
+            }
+        }
+
         public Operation(IntPtr handle)
         {
             if (handle == IntPtr.Zero)
@@ -195,7 +207,7 @@ namespace Tensorflow
             return c_api.TF_OperationGetAttrMetadata(_handle, attr_name, s);
         }
 
-        public NodeDef GetNodeDef()
+        private NodeDef GetNodeDef()
         {
             using (var s = new Status())
             using (var buffer = new Buffer())

src/TensorFlowNET.Core/Tensors/TF_DataType.cs

@@ -36,6 +36,8 @@ namespace Tensorflow
         TF_UINT32 = 22,
         TF_UINT64 = 23,
 
+        DtFloatRef = 101, // DT_FLOAT_REF
         DtDoubleRef = 102, // DT_DOUBLE_REF
+        DtInt32Ref = 103, // DT_INT32_REF
     }
 }

src/TensorFlowNET.Core/Tensors/Tensor.cs

@@ -162,6 +162,7 @@ namespace Tensorflow
             this.op = op;
             this.value_index = value_index;
             this._dtype = dtype;
+            _id = ops.uid();
         }
 
         public List<Operation> consumers()

src/TensorFlowNET.Core/Variables/RefVariable.cs

@@ -1,5 +1,6 @@
 using System;
 using System.Collections.Generic;
+using System.Linq;
 using System.Text;
 
 namespace Tensorflow
@@ -69,14 +70,15 @@ namespace Tensorflow
                 {
 
                 }
+                // Or get the initial value from a Tensor or Python object.
                 else
                 {
                     _initial_value = ops.convert_to_tensor(initial_value, name: "initial_value");
-                }
 
-                var shape = _initial_value.shape;
-                dtype = _initial_value.dtype;
-                _variable = gen_state_ops.variable_v2(shape, dtype, name);
+                    var shape = _initial_value.shape;
+                    dtype = _initial_value.dtype;
+                    _variable = gen_state_ops.variable_v2(shape, dtype, name);
+                }
 
                 // Manually overrides the variable's shape with the initial value's.
                 if (validate_shape)
@@ -87,8 +89,9 @@ namespace Tensorflow
                 // If 'initial_value' makes use of other variables, make sure we don't
                 // have an issue if these other variables aren't initialized first by
                 // using their initialized_value() method.
+                var _initial_value2 = _try_guard_against_uninitialized_dependencies(_initial_value);
 
-                _initializer_op = gen_state_ops.assign(_variable, _initial_value, validate_shape).op;
+                _initializer_op = gen_state_ops.assign(_variable, _initial_value2, validate_shape).op;
 
                 if (!String.IsNullOrEmpty(caching_device))
                 {
@@ -112,5 +115,51 @@ namespace Tensorflow
         {
             return _snapshot;
         }
+
+        /// <summary>
+        /// Attempt to guard against dependencies on uninitialized variables.
+        /// </summary>
+        /// <param name="initial_value"></param>
+        private Tensor _try_guard_against_uninitialized_dependencies(Tensor initial_value)
+        {
+            return _safe_initial_value_from_tensor(initial_value, new Dictionary<string, Operation>());
+        }
+
+        /// <summary>
+        /// Replace dependencies on variables with their initialized values.
+        /// </summary>
+        /// <param name="tensor">A `Tensor`. The tensor to replace.</param>
+        /// <param name="op_cache">A dict mapping operation names to `Operation`s.</param>
+        /// <returns>A `Tensor` compatible with `tensor`.</returns>
+        private Tensor _safe_initial_value_from_tensor(Tensor tensor, Dictionary<string, Operation> op_cache)
+        {
+            var op = tensor.op;
+            var new_op = op_cache.ContainsKey(op.Name) ? op_cache[op.Name] : null;
+            if(new_op == null)
+            {
+                new_op = _safe_initial_value_from_op(op, op_cache);
+                op_cache[op.Name] = new_op;
+            }
+            return new_op.outputs[tensor.value_index];
+        }
+
+        private Operation _safe_initial_value_from_op(Operation op, Dictionary<string, Operation> op_cache)
+        {
+            var op_type = op.node_def.Op;
+            switch (op_type)
+            {
+                case "IsVariableInitialized":
+                case "VarIsInitializedOp":
+                case "ReadVariableOp":
+                    return op;
+                case "Variable":
+                case "VariableV2":
+                case "VarHandleOp":
+                    break;
+            }
+
+            // Recursively build initializer expressions for inputs.
+            return op;
+        }
     }
 }

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

@@ -42,7 +42,7 @@ namespace Tensorflow
 
             _execute.record_gradient("VariableV2", _inputs_flat, _attrs, _result, name);
 
-            return new Tensor(_op, 0, dtype);
+            return _result[0];
         }
 
         /// <summary>

test/TensorFlowNET.UnitTest/GraphTest.cs

@@ -130,7 +130,7 @@ namespace TensorFlowNET.UnitTest
             EXPECT_EQ(TF_Code.TF_OK, s.Code);
 
             // Serialize to NodeDef.
-            var node_def = neg.GetNodeDef();
+            var node_def = neg.node_def;
 
             // Validate NodeDef is what we expect.
             ASSERT_TRUE(c_test_util.IsNeg(node_def, "add"));
@@ -145,13 +145,13 @@ namespace TensorFlowNET.UnitTest
             // Look up some nodes by name.
             Operation neg2 = c_api.TF_GraphOperationByName(graph, "neg");
             EXPECT_EQ(neg, neg2);
-            var node_def2 = neg2.GetNodeDef();
+            var node_def2 = neg2.node_def;
             EXPECT_EQ(node_def.ToString(), node_def2.ToString());
 
             Operation feed2 = c_api.TF_GraphOperationByName(graph, "feed");
             EXPECT_EQ(feed, feed2);
-            node_def = feed.GetNodeDef();
-            node_def2 = feed2.GetNodeDef();
+            node_def = feed.node_def;
+            node_def2 = feed2.node_def;
             EXPECT_EQ(node_def.ToString(), node_def2.ToString());
 
             // Test iterating through the nodes of a graph.
@@ -186,7 +186,7 @@ namespace TensorFlowNET.UnitTest
                 }
                 else
                 {
-                    node_def = oper.GetNodeDef();
+                    node_def = oper.node_def;
                     Assert.Fail($"Unexpected Node: {node_def.ToString()}");
                 }
             }