Automatically add KerasInterface to f.

src/TensorFlowNET.Console/Program.cs

@@ -8,8 +8,6 @@ namespace Tensorflow
     {
         static void Main(string[] args)
         {
-            tf.UseKeras<KerasInterface>();
-
             var diag = new Diagnostician();
             // diag.Diagnose(@"D:\memory.txt");
 

src/TensorFlowNET.Core/Keras/IKerasApi.cs

@@ -1,19 +1,24 @@
 using System;
 using System.Collections.Generic;
 using System.Text;
+using System.Threading;
+using Tensorflow.Framework.Models;
 using Tensorflow.Keras.Engine;
 using Tensorflow.Keras.Layers;
 using Tensorflow.Keras.Losses;
 using Tensorflow.Keras.Metrics;
+using Tensorflow.Keras.Models;
 
 namespace Tensorflow.Keras
 {
     public interface IKerasApi
     {
-        public ILayersApi layers { get; }
-        public ILossesApi losses { get; }
-        public IMetricsApi metrics { get; }
-        public IInitializersApi initializers { get; }
+        IInitializersApi initializers { get; }
+        ILayersApi layers { get; }
+        ILossesApi losses { get; }
+        IOptimizerApi optimizers { get; }
+        IMetricsApi metrics { get; }
+        IModelsApi models { get; }
 
         /// <summary>
         /// `Model` groups layers into an object with training and inference features.
@@ -21,6 +26,35 @@ namespace Tensorflow.Keras
         /// <param name="input"></param>
         /// <param name="output"></param>
         /// <returns></returns>
-        public IModel Model(Tensors inputs, Tensors outputs, string name = null);
+        IModel Model(Tensors inputs, Tensors outputs, string name = null);
+
+        /// <summary>
+        /// Instantiate a Keras tensor.
+        /// </summary>
+        /// <param name="shape"></param>
+        /// <param name="batch_size"></param>
+        /// <param name="dtype"></param>
+        /// <param name="name"></param>
+        /// <param name="sparse">
+        /// A boolean specifying whether the placeholder to be created is sparse.
+        /// </param>
+        /// <param name="ragged">
+        /// A boolean specifying whether the placeholder to be created is ragged.
+        /// </param>
+        /// <param name="tensor">
+        /// Optional existing tensor to wrap into the `Input` layer.
+        /// If set, the layer will not create a placeholder tensor.
+        /// </param>
+        /// <returns></returns>
+        Tensors Input(Shape shape = null,
+            int batch_size = -1,
+            string name = null,
+            TF_DataType dtype = TF_DataType.DtInvalid,
+            bool sparse = false,
+            Tensor tensor = null,
+            bool ragged = false,
+            TypeSpec type_spec = null,
+            Shape batch_input_shape = null,
+            Shape batch_shape = null);
     }
 }

src/TensorFlowNET.Core/Keras/IOptimizerApi.cs

@@ -0,0 +1,47 @@
+using System;
+using System.Collections.Generic;
+using System.Text;
+using Tensorflow.Keras.Engine;
+
+namespace Tensorflow.Keras
+{
+    public interface IOptimizerApi
+    {
+        /// <summary>
+        /// Adam optimization is a stochastic gradient descent method that is based on
+        /// adaptive estimation of first-order and second-order moments.
+        /// </summary>
+        /// <param name="learning_rate"></param>
+        /// <param name="beta_1"></param>
+        /// <param name="beta_2"></param>
+        /// <param name="epsilon"></param>
+        /// <param name="amsgrad"></param>
+        /// <param name="name"></param>
+        /// <returns></returns>
+        IOptimizer Adam(float learning_rate = 0.001f,
+                float beta_1 = 0.9f,
+                float beta_2 = 0.999f,
+                float epsilon = 1e-7f,
+                bool amsgrad = false,
+                string name = "Adam");
+
+        /// <summary>
+        /// Construct a new RMSprop optimizer.
+        /// </summary>
+        /// <param name="learning_rate"></param>
+        /// <param name="rho"></param>
+        /// <param name="momentum"></param>
+        /// <param name="epsilon"></param>
+        /// <param name="centered"></param>
+        /// <param name="name"></param>
+        /// <returns></returns>
+        IOptimizer RMSprop(float learning_rate = 0.001f,
+                float rho = 0.9f,
+                float momentum = 0.0f,
+                float epsilon = 1e-7f,
+                bool centered = false,
+                string name = "RMSprop");
+
+        IOptimizer SGD(float learning_rate);
+    }
+}

src/TensorFlowNET.Core/Keras/Models/IModelsApi.cs

@@ -0,0 +1,12 @@
+using System;
+using System.Collections.Generic;
+using System.Text;
+using Tensorflow.Keras.Engine;
+
+namespace Tensorflow.Keras.Models
+{
+    public interface IModelsApi
+    {
+        public IModel load_model(string filepath, bool compile = true, LoadOptions? options = null);
+    }
+}

src/TensorFlowNET.Core/tensorflow.cs

@@ -65,14 +65,6 @@ namespace Tensorflow
             InitGradientEnvironment();
         }
 
-        public void UseKeras<T>() where T : IKerasApi, new()
-        {
-            if (keras == null)
-            {
-                keras = new T();
-            }
-        }
-
         public string VERSION => c_api.StringPiece(c_api.TF_Version());
 
         private void InitGradientEnvironment()

src/TensorFlowNET.Keras/KerasApi.cs

@@ -7,6 +7,6 @@ namespace Tensorflow
     /// </summary>
     public static class KerasApi
     {
-        public static KerasInterface keras { get; } = new KerasInterface();
+        public static KerasInterface keras { get; } = KerasInterface.Instance;
     }
 }

src/TensorFlowNET.Keras/KerasInterface.cs

@@ -18,6 +18,28 @@ namespace Tensorflow.Keras
 {
     public class KerasInterface : IKerasApi
     {
+        private static KerasInterface _instance = null;
+        private static readonly object _lock = new object();  
+        private KerasInterface()
+        {
+            Tensorflow.Binding.tf.keras = this;
+        }
+
+        public static KerasInterface Instance
+        {
+            get
+            {
+                lock (_lock)
+                {
+                    if (_instance is null)
+                    {
+                        _instance = new KerasInterface();
+                    }
+                    return _instance;
+                }
+            }
+        }
+
         public KerasDataset datasets { get; } = new KerasDataset();
         public IInitializersApi initializers { get; } = new InitializersApi();
         public Regularizers regularizers { get; } = new Regularizers();
@@ -27,9 +49,9 @@ namespace Tensorflow.Keras
         public Preprocessing preprocessing { get; } = new Preprocessing();
         ThreadLocal<BackendImpl> _backend = new ThreadLocal<BackendImpl>(() => new BackendImpl());
         public BackendImpl backend => _backend.Value;
-        public OptimizerApi optimizers { get; } = new OptimizerApi();
+        public IOptimizerApi optimizers { get; } = new OptimizerApi();
         public IMetricsApi metrics { get; } = new MetricsApi();
-        public ModelsApi models { get; } = new ModelsApi();
+        public IModelsApi models { get; } = new ModelsApi();
         public KerasUtils utils { get; } = new KerasUtils();
 
         public Sequential Sequential(List<ILayer> layers = null,

src/TensorFlowNET.Keras/Models/ModelsApi.cs

@@ -9,12 +9,12 @@ using ThirdParty.Tensorflow.Python.Keras.Protobuf;
 
 namespace Tensorflow.Keras.Models
 {
-    public class ModelsApi
+    public class ModelsApi: IModelsApi
     {
         public Functional from_config(ModelConfig config)
             => Functional.from_config(config);
 
-        public Model load_model(string filepath, bool compile = true, LoadOptions? options = null)
+        public IModel load_model(string filepath, bool compile = true, LoadOptions? options = null)
         {
             return KerasLoadModelUtils.load_model(filepath, compile: compile, options: options) as Model;
         }

src/TensorFlowNET.Keras/Optimizers/OptimizerApi.cs

@@ -1,8 +1,9 @@
 using Tensorflow.Keras.ArgsDefinition;
+using Tensorflow.Keras.Engine;
 
 namespace Tensorflow.Keras.Optimizers
 {
-    public class OptimizerApi
+    public class OptimizerApi: IOptimizerApi
     {
         /// <summary>
         /// Adam optimization is a stochastic gradient descent method that is based on
@@ -15,7 +16,7 @@ namespace Tensorflow.Keras.Optimizers
         /// <param name="amsgrad"></param>
         /// <param name="name"></param>
         /// <returns></returns>
-        public OptimizerV2 Adam(float learning_rate = 0.001f,
+        public IOptimizer Adam(float learning_rate = 0.001f,
                 float beta_1 = 0.9f,
                 float beta_2 = 0.999f,
                 float epsilon = 1e-7f,
@@ -38,7 +39,7 @@ namespace Tensorflow.Keras.Optimizers
         /// <param name="centered"></param>
         /// <param name="name"></param>
         /// <returns></returns>
-        public OptimizerV2 RMSprop(float learning_rate = 0.001f,
+        public IOptimizer RMSprop(float learning_rate = 0.001f,
                 float rho = 0.9f,
                 float momentum = 0.0f,
                 float epsilon = 1e-7f,
@@ -54,7 +55,7 @@ namespace Tensorflow.Keras.Optimizers
                 Name = name
             });
 
-        public SGD SGD(float learning_rate)
+        public IOptimizer SGD(float learning_rate)
             => new SGD(learning_rate);
     }
 }

test/TensorFlowNET.Keras.UnitTest/EagerModeTestBase.cs

@@ -10,8 +10,6 @@ namespace TensorFlowNET.Keras.UnitTest
         [TestInitialize]
         public void TestInit()
         {
-            tf.UseKeras<KerasInterface>();
-
             if (!tf.executing_eagerly())
                 tf.enable_eager_execution();
             tf.Context.ensure_initialized();

test/TensorFlowNET.Keras.UnitTest/Layers/LayersTest.cs

@@ -150,7 +150,6 @@ namespace TensorFlowNET.Keras.UnitTest
         [TestMethod, Ignore("WIP")]
         public void SimpleRNN()
         {
-            tf.UseKeras<KerasInterface>();
             var inputs = np.arange(6 * 10 * 8).reshape((6, 10, 8)).astype(np.float32);
             /*var simple_rnn = keras.layers.SimpleRNN(4);
             var output = simple_rnn.Apply(inputs);

test/TensorFlowNET.Keras.UnitTest/Layers/ModelSaveTest.cs

@@ -3,6 +3,7 @@ using Tensorflow.Keras.Engine;
 using System.Diagnostics;
 using static Tensorflow.KerasApi;
 using Tensorflow.Keras.Saving;
+using Tensorflow.Keras.Models;
 
 namespace TensorFlowNET.Keras.UnitTest
 {
@@ -18,7 +19,7 @@ namespace TensorFlowNET.Keras.UnitTest
             var model = GetFunctionalModel();
             var config = model.get_config();
             Debug.Assert(config is ModelConfig);
-            var new_model = keras.models.from_config(config as ModelConfig);
+            var new_model = new ModelsApi().from_config(config as ModelConfig);
             Assert.AreEqual(model.Layers.Count, new_model.Layers.Count);
         }