Sequential #570

Author: Oceania2018

src/TensorFlowNET.Core/APIs/tf.init.cs

@@ -27,7 +27,8 @@ public IInitializer constant_initializer<T>(T value, TF_DataType dtype = TF_Data
         public IInitializer zeros_initializer => new Zeros();
         public IInitializer ones_initializer => new Ones();
         public IInitializer glorot_uniform_initializer => new GlorotUniform();
-        public IInitializer uniform_initializer => new RandomUniform();
+        public IInitializer random_uniform_initializer => new RandomUniform();
+        public IInitializer orthogonal_initializer => new Orthogonal();
 
         public variable_scope variable_scope(string name,
                string default_name = null,

src/TensorFlowNET.Core/Keras/Activations/Activations.Relu.cs

@@ -20,7 +20,9 @@ public partial class Activations
                 return results[0];
             }
 
-            throw new NotImplementedException("");
+            var _op = tf.OpDefLib._apply_op_helper("Relu", name: name, args: new { features });
+
+            return _op.output;
         };
     }
 }

src/TensorFlowNET.Core/Keras/Activations/Activations.Sigmoid.cs

@@ -0,0 +1,26 @@
+using System;
+using System.Collections.Generic;
+using System.Text;
+using Tensorflow.Operations;
+using static Tensorflow.Binding;
+
+namespace Tensorflow.Keras
+{
+    public partial class Activations
+    {
+        public Activation Sigmoid = (features, name) =>
+        {
+            if (tf.executing_eagerly())
+            {
+                var results = tf.Runner.TFE_FastPathExecute(tf.Context, tf.Context.DeviceName,
+                    "Sigmoid", name,
+                    null,
+                    features);
+
+                return results[0];
+            }
+
+            throw new NotImplementedException("");
+        };
+    }
+}

src/TensorFlowNET.Core/Keras/Activations/Activations.Tanh.cs

@@ -0,0 +1,26 @@
+using System;
+using System.Collections.Generic;
+using System.Text;
+using Tensorflow.Operations;
+using static Tensorflow.Binding;
+
+namespace Tensorflow.Keras
+{
+    public partial class Activations
+    {
+        public Activation Tanh = (features, name) =>
+        {
+            if (tf.executing_eagerly())
+            {
+                var results = tf.Runner.TFE_FastPathExecute(tf.Context, tf.Context.DeviceName,
+                    "Tanh", name,
+                    null,
+                    features);
+
+                return results[0];
+            }
+
+            throw new NotImplementedException("");
+        };
+    }
+}

src/TensorFlowNET.Core/Keras/ArgsDefinition/EmbeddingArgs.cs

@@ -0,0 +1,15 @@
+using System;
+using System.Collections.Generic;
+using System.Text;
+
+namespace Tensorflow.Keras.ArgsDefinition
+{
+    public class EmbeddingArgs : LayerArgs
+    {
+        public int InputDim { get; set; }
+        public int OutputDim { get; set; }
+        public bool MaskZero { get; set; }
+        public int InputLength { get; set; } = -1;
+        public IInitializer EmbeddingsInitializer { get; set; }
+    }
+}

src/TensorFlowNET.Core/Keras/ArgsDefinition/LSTMArgs.cs

@@ -0,0 +1,26 @@
+using System;
+using System.Collections.Generic;
+using System.Text;
+
+namespace Tensorflow.Keras.ArgsDefinition
+{
+    public class LSTMArgs : RNNArgs
+    {
+        public int Units { get; set; }
+        public Activation Activation { get; set; }
+        public Activation RecurrentActivation { get; set; }
+        public IInitializer KernelInitializer { get; set; }
+        public IInitializer RecurrentInitializer { get; set; }
+        public IInitializer BiasInitializer { get; set; }
+        public bool UnitForgetBias { get; set; }
+        public float Dropout { get; set; }
+        public float RecurrentDropout { get; set; }
+        public int Implementation { get; set; }
+        public bool ReturnSequences { get; set; }
+        public bool ReturnState { get; set; }
+        public bool GoBackwards { get; set; }
+        public bool Stateful { get; set; }
+        public bool TimeMajor { get; set; }
+        public bool Unroll { get; set; }
+    }
+}

src/TensorFlowNET.Core/Keras/ArgsDefinition/LSTMCellArgs.cs

@@ -0,0 +1,10 @@
+using System;
+using System.Collections.Generic;
+using System.Text;
+
+namespace Tensorflow.Keras.ArgsDefinition
+{
+    public class LSTMCellArgs : LayerArgs
+    {
+    }
+}

src/TensorFlowNET.Core/Keras/ArgsDefinition/RNNArgs.cs

@@ -0,0 +1,10 @@
+using System;
+using System.Collections.Generic;
+using System.Text;
+
+namespace Tensorflow.Keras.ArgsDefinition
+{
+    public class RNNArgs : LayerArgs
+    {
+    }
+}

src/TensorFlowNET.Core/Keras/Engine/InputSpec.cs

@@ -26,17 +26,22 @@ public class InputSpec
         public int? ndim;
         public int? min_ndim;
         Dictionary<int, int> axes;
+        TensorShape shape;
 
         public InputSpec(TF_DataType dtype = TF_DataType.DtInvalid,
             int? ndim = null,
             int? min_ndim = null,
-            Dictionary<int, int> axes = null)
+            Dictionary<int, int> axes = null,
+            TensorShape shape = null)
         {
             this.ndim = ndim;
             if (axes == null)
                 axes = new Dictionary<int, int>();
             this.axes = axes;
             this.min_ndim = min_ndim;
+            this.shape = shape;
+            if (ndim == null && shape != null)
+                this.ndim = shape.ndim;
         }
     }
 }

src/TensorFlowNET.Core/Keras/Engine/Layer.Layers.cs

@@ -3,6 +3,7 @@
 using System.Text;
 using Tensorflow.Keras.ArgsDefinition;
 using Tensorflow.Keras.Layers;
+using Tensorflow.Operations.Activation;
 using static Tensorflow.Binding;
 
 namespace Tensorflow.Keras.Engine
@@ -100,5 +101,46 @@ protected Layer Flatten()
             _layers.Add(layer);
             return layer;
         }
+
+        protected Layer LSTM(int units,
+            Activation activation = null,
+            Activation recurrent_activation = null,
+            bool use_bias = true,
+            IInitializer kernel_initializer = null,
+            IInitializer recurrent_initializer = null,
+            IInitializer bias_initializer = null,
+            bool unit_forget_bias = true,
+            float dropout = 0f,
+            float recurrent_dropout = 0f,
+            int implementation = 2,
+            bool return_sequences = false,
+            bool return_state = false,
+            bool go_backwards = false,
+            bool stateful = false,
+            bool time_major = false,
+            bool unroll = false)
+        {
+            var layer = new LSTM(new LSTMArgs
+            {
+                Units = units,
+                Activation = activation ?? tf.keras.activations.Tanh,
+                RecurrentActivation = recurrent_activation ?? tf.keras.activations.Sigmoid,
+                KernelInitializer = kernel_initializer ?? tf.glorot_uniform_initializer,
+                RecurrentInitializer = recurrent_initializer ?? tf.orthogonal_initializer,
+                BiasInitializer = bias_initializer ?? tf.zeros_initializer,
+                Dropout = dropout,
+                RecurrentDropout = recurrent_dropout,
+                Implementation = implementation,
+                ReturnSequences = return_sequences,
+                ReturnState = return_state,
+                GoBackwards = go_backwards,
+                Stateful = stateful,
+                TimeMajor = time_major,
+                Unroll = unroll
+            });
+
+            _layers.Add(layer);
+            return layer;
+        }
     }
 }