Support mutiple inputs of keras modek.predict.

Author: AsakusaRinne

src/TensorFlowNET.Core/Data/OwnedIterator.cs

@@ -13,7 +13,7 @@ public class OwnedIterator : IDisposable
         IDatasetV2 _dataset;
         TensorSpec[] _element_spec;
         dataset_ops ops = new dataset_ops();
-        Tensor _deleter;
+        //Tensor _deleter;
         Tensor _iterator_resource;
 
         public OwnedIterator(IDatasetV2 dataset)
@@ -26,7 +26,6 @@ void _create_iterator(IDatasetV2 dataset)
             dataset = dataset.apply_options();
             _dataset = dataset;
             _element_spec = dataset.element_spec;
-            // _flat_output_types = 
             _iterator_resource = ops.anonymous_iterator_v3(_dataset.output_types, _dataset.output_shapes);
             // TODO(Rinne): deal with graph mode.
             ops.make_iterator(dataset.variant_tensor, _iterator_resource);

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

@@ -62,7 +62,7 @@ void evaluate(NDArray x, NDArray y,
             bool use_multiprocessing = false,
             bool return_dict = false);
 
-    Tensors predict(Tensor x,
+    Tensors predict(Tensors x,
             int batch_size = -1,
             int verbose = 0,
             int steps = -1,

src/TensorFlowNET.Keras/Engine/Model.Predict.cs

@@ -49,7 +49,7 @@ public Tensors predict(IDatasetV2 dataset,
         /// <param name="workers"></param>
         /// <param name="use_multiprocessing"></param>
         /// <returns></returns>
-        public Tensors predict(Tensor x,
+        public Tensors predict(Tensors x,
             int batch_size = -1,
             int verbose = 0,
             int steps = -1,
@@ -115,12 +115,12 @@ Tensors PredictInternal(DataHandler data_handler, int verbose)
         Tensors run_predict_step(OwnedIterator iterator)
         {
             var data = iterator.next();
-            var outputs = predict_step(data[0]);
+            var outputs = predict_step(data);
             tf_with(ops.control_dependencies(new object[0]), ctl => _predict_counter.assign_add(1));
             return outputs;
         }
 
-        Tensors predict_step(Tensor data)
+        Tensors predict_step(Tensors data)
         {
             return Apply(data, training: false);
         }

test/TensorFlowNET.Keras.UnitTest/MultiInputModelTest.cs

@@ -1,27 +1,17 @@
-using Microsoft.VisualStudio.TestPlatform.Utilities;
-using Microsoft.VisualStudio.TestTools.UnitTesting;
+using Microsoft.VisualStudio.TestTools.UnitTesting;
 using System;
-using System.Collections.Generic;
-using System.Linq;
-using System.Text;
-using System.Threading.Tasks;
-using System.Xml.Linq;
-using Tensorflow.Operations;
-using static Tensorflow.Binding;
-using static Tensorflow.KerasApi;
-using Tensorflow.NumPy;
-using Microsoft.VisualBasic;
-using static HDF.PInvoke.H5T;
-using Tensorflow.Keras.UnitTest.Helpers;
+using Tensorflow;
 using Tensorflow.Keras.Optimizers;
+using Tensorflow.NumPy;
+using static Tensorflow.KerasApi;
 
-namespace Tensorflow.Keras.UnitTest
+namespace TensorFlowNET.Keras.UnitTest
 {
     [TestClass]
     public class MultiInputModelTest
     {
         [TestMethod]
-        public void SimpleModel()
+        public void LeNetModel()
         {
             var inputs = keras.Input((28, 28, 1));
             var conv1 = keras.layers.Conv2D(16, (3, 3), activation: "relu", padding: "same").Apply(inputs);
@@ -40,7 +30,7 @@ public void SimpleModel()
             var concat = keras.layers.Concatenate().Apply((flat1, flat1_2));
             var dense1 = keras.layers.Dense(512, activation: "relu").Apply(concat);
             var dense2 = keras.layers.Dense(128, activation: "relu").Apply(dense1);
-            var dense3 = keras.layers.Dense(10, activation:  "relu").Apply(dense2);
+            var dense3 = keras.layers.Dense(10, activation: "relu").Apply(dense2);
             var output = keras.layers.Softmax(-1).Apply(dense3);
 
             var model = keras.Model((inputs, inputs_2), output);
@@ -52,7 +42,7 @@ public void SimpleModel()
             {
                 TrainDir = "mnist",
                 OneHot = false,
-                ValidationSize = 59000,
+                ValidationSize = 59900,
             }).Result;
 
             var loss = keras.losses.SparseCategoricalCrossentropy();
@@ -64,6 +54,11 @@ public void SimpleModel()
 
             var x = new NDArray[] { x1, x2 };
             model.fit(x, dataset.Train.Labels, batch_size: 8, epochs: 3);
+
+            x1 = np.ones((1, 28, 28, 1), TF_DataType.TF_FLOAT);
+            x2 = np.zeros((1, 28, 28, 1), TF_DataType.TF_FLOAT);
+            var pred = model.predict((x1, x2));
+            Console.WriteLine(pred);
         }
     }
 }