Add ZipDataset, fix paths_and_labels_to_dataset.

Author: Oceania2018

src/TensorFlowNET.Core/Data/DatasetManager.cs

@@ -25,10 +25,16 @@ public IDatasetV2 from_tensor(Tensor tensors)
         public IDatasetV2 from_tensor_slices(Tensor features, Tensor labels)
             => new TensorSliceDataset(features, labels);
 
+        public IDatasetV2 from_tensor_slices(NDArray array)
+            => new TensorSliceDataset(array);
+
         public IDatasetV2 range(int count, TF_DataType output_type = TF_DataType.TF_INT64)
             => new RangeDataset(count, output_type: output_type);
 
         public IDatasetV2 range(int start, int stop, int step = 1, TF_DataType output_type = TF_DataType.TF_INT64)
             => new RangeDataset(stop, start: start, step: step, output_type: output_type);
+
+        public IDatasetV2 zip(params IDatasetV2[] ds)
+            => new ZipDataset(ds);
     }
 }

src/TensorFlowNET.Core/Data/TensorSliceDataset.cs

@@ -11,6 +11,16 @@ namespace Tensorflow.Data
 {
     public class TensorSliceDataset : DatasetSource
     {
+        public TensorSliceDataset(NDArray array)
+        {
+            var element = tf.constant(array);
+            _tensors = new[] { element };
+            var batched_spec = new[] { element.ToTensorSpec() };
+            structure = batched_spec.Select(x => x._unbatch()).ToArray();
+
+            variant_tensor = ops.tensor_slice_dataset(_tensors, output_shapes);
+        }
+
         public TensorSliceDataset(Tensor features, Tensor labels)
         {
             _tensors = new[] { features, labels };

src/TensorFlowNET.Core/Data/ZipDataset.cs

@@ -0,0 +1,18 @@
+using System;
+using System.Collections.Generic;
+using System.Linq;
+using System.Text;
+
+namespace Tensorflow
+{
+    public class ZipDataset : DatasetV2
+    {
+        dataset_ops ops = new dataset_ops();
+        public ZipDataset(params IDatasetV2[] ds)
+        {
+            var input_datasets = ds.Select(x => x.variant_tensor).ToArray();
+            structure = ds.Select(x => x.structure[0]).ToArray();
+            variant_tensor = ops.zip_dataset(input_datasets, output_types, output_shapes);
+        }
+    }
+}

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

@@ -0,0 +1,12 @@
+using System;
+using System.Collections.Generic;
+using System.Text;
+
+namespace Tensorflow.Keras.ArgsDefinition
+{
+    public class RescalingArgs : LayerArgs
+    {
+        public float Scale { get; set; }
+        public float Offset { get; set; }
+    }
+}

src/TensorFlowNET.Core/Keras/KerasApi.cs

@@ -72,7 +72,17 @@ public Tensor Input(TensorShape shape = null,
 
         public class LayersApi
         {
-            public Layer Dense(int units,
+            public Rescaling Rescaling(float scale,
+                float offset = 0,
+                TensorShape input_shape = null)
+                => new Rescaling(new RescalingArgs
+                {
+                    Scale = scale,
+                    Offset = offset,
+                    InputShape = input_shape
+                });
+
+            public Dense Dense(int units,
                 Activation activation = null,
                 TensorShape input_shape = null)
                 => new Dense(new DenseArgs

src/TensorFlowNET.Core/Keras/Layers/Rescaling.cs

@@ -0,0 +1,29 @@
+using System;
+using System.Collections.Generic;
+using System.Text;
+using Tensorflow.Keras.ArgsDefinition;
+using Tensorflow.Keras.Engine;
+
+namespace Tensorflow.Keras.Layers
+{
+    /// <summary>
+    /// Multiply inputs by `scale` and adds `offset`.
+    /// </summary>
+    public class Rescaling : Layer
+    {
+        RescalingArgs args;
+        Tensor scale;
+        Tensor offset;
+
+        public Rescaling(RescalingArgs args) : base(args)
+        {
+            this.args = args;
+        }
+
+        protected override Tensor call(Tensor inputs, bool is_training = false, Tensor state = null)
+        {
+            scale = math_ops.cast(args.Scale, args.DType);
+            throw new NotImplementedException("");
+        }
+    }
+}

src/TensorFlowNET.Core/Keras/Preprocessings/DatasetUtils.cs

@@ -1,11 +1,20 @@
 using System;
 using System.Collections.Generic;
 using System.Text;
+using static Tensorflow.Binding;
 
 namespace Tensorflow.Keras.Preprocessings
 {
     public partial class DatasetUtils
     {
-
+        public IDatasetV2 labels_to_dataset(int[] labels, string label_mode, int num_classes)
+        {
+            var label_ds = tf.data.Dataset.from_tensor_slices(labels);
+            if (label_mode == "binary")
+                throw new NotImplementedException("");
+            else if(label_mode == "categorical")
+                throw new NotImplementedException("");
+            return label_ds;
+        }
     }
 }

src/TensorFlowNET.Core/Keras/Preprocessings/DatasetUtils.get_training_or_validation_split.cs

@@ -31,8 +31,8 @@ public partial class DatasetUtils
             else if (subset == "validation")
             {
                 Console.WriteLine($"Using {num_val_samples} files for validation.");
-                samples = samples[samples.Length..];
-                labels = labels[samples.Length..];
+                samples = samples[(samples.Length - num_val_samples)..];
+                labels = labels[(labels.Length - num_val_samples)..];
             }
             else
                 throw new NotImplementedException("");

src/TensorFlowNET.Core/Keras/Preprocessings/Preprocessing.image_dataset_from_directory.cs

@@ -25,7 +25,7 @@ public partial class Preprocessing
         /// <param name="interpolation"></param>
         /// <param name="follow_links"></param>
         /// <returns></returns>
-        public Tensor image_dataset_from_directory(string directory,
+        public IDatasetV2 image_dataset_from_directory(string directory,
             string labels = "inferred",
             string label_mode = "int",
             string[] class_names = null,
@@ -52,8 +52,11 @@ public Tensor image_dataset_from_directory(string directory,
 
             (image_paths, label_list) = tf.keras.preprocessing.dataset_utils.get_training_or_validation_split(image_paths, label_list, validation_split, subset);
 
-            paths_and_labels_to_dataset(image_paths, image_size, num_channels, label_list, label_mode, class_name_list.Length, interpolation);
-            throw new NotImplementedException("");
+            var dataset = paths_and_labels_to_dataset(image_paths, image_size, num_channels, label_list, label_mode, class_name_list.Length, interpolation);
+            if (shuffle)
+                dataset = dataset.shuffle(batch_size * 8, seed: seed);
+            dataset = dataset.batch(batch_size);
+            return dataset;
         }
     }
 }

src/TensorFlowNET.Core/Keras/Preprocessings/Preprocessing.paths_and_labels_to_dataset.cs

@@ -1,23 +1,44 @@
-using System;
+using NumSharp;
+using System;
 using System.Globalization;
+using System.Threading.Tasks;
 using static Tensorflow.Binding;
 
 namespace Tensorflow.Keras
 {
     public partial class Preprocessing
     {
-        public Tensor paths_and_labels_to_dataset(string[] image_paths,
+        public IDatasetV2 paths_and_labels_to_dataset(string[] image_paths,
             TensorShape image_size,
             int num_channels,
             int[] labels,
             string label_mode,
             int num_classes,
             string interpolation)
         {
-            foreach (var image_path in image_paths)
-                path_to_image(image_path, image_size, num_channels, interpolation);
+            Shape shape = (image_paths.Length, image_size.dims[0], image_size.dims[1], num_channels);
+            Console.WriteLine($"Allocating memory for shape{shape}, {NPTypeCode.Float}");
+            var data = np.zeros(shape, NPTypeCode.Float);
 
-            throw new NotImplementedException("");
+            for (var i = 0; i < image_paths.Length; i++)
+            {
+                var image = path_to_image(image_paths[i], image_size, num_channels, interpolation);
+                data[i] = image.numpy();
+                if (i % 100 == 0)
+                    Console.WriteLine($"Filled {i}/{image_paths.Length} data into memory.");
+            }
+
+            var img_ds = tf.data.Dataset.from_tensor_slices(data);
+
+            if (label_mode == "int")
+            {
+                var label_ds = tf.keras.preprocessing.dataset_utils.labels_to_dataset(labels, label_mode, num_classes);
+                img_ds = tf.data.Dataset.zip(img_ds, label_ds);
+            }
+            else
+                throw new NotImplementedException("");
+
+            return img_ds;
         }
 
         Tensor path_to_image(string path, TensorShape image_size, int num_channels, string interpolation)