Merge pull request #1140 from dogvane/master

fix same bug

Author: Oceania2018

src/TensorFlowNET.Core/Gradients/nn_grad.cs

@@ -365,6 +365,23 @@ public static Tensor[] _MaxPoolGrad(Operation op, Tensor[] grads)
             };
         }
 
+        [RegisterGradient("AvgPool")]
+        public static Tensor[] _AvgPoolGrad(Operation op, Tensor[] grads)
+        {
+            Tensor grad = grads[0];
+            
+            return new Tensor[]
+            {
+                gen_nn_ops.avg_pool_grad(
+                  array_ops.shape(op.inputs[0]),
+                  grad,
+                  op.get_attr_list<int>("ksize"),
+                  op.get_attr_list<int>("strides"),
+                  op.get_attr("padding").ToString(),
+                  op.get_attr("data_format").ToString())
+            };
+        }
+
         /// <summary>
         /// Return the gradients for TopK.
         /// </summary>

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

@@ -15,7 +15,7 @@ public interface ILayer: IWithTrackable, IKerasConfigable
         List<ILayer> Layers { get; }
         List<INode> InboundNodes { get; }
         List<INode> OutboundNodes { get; }
-        Tensors Apply(Tensors inputs, Tensors states = null, bool training = false, IOptionalArgs? optional_args = null);
+        Tensors Apply(Tensors inputs, Tensors states = null, bool? training = false, IOptionalArgs? optional_args = null);
         List<IVariableV1> TrainableVariables { get; }
         List<IVariableV1> TrainableWeights { get; }
         List<IVariableV1> NonTrainableWeights { get; }

src/TensorFlowNET.Core/Operations/NnOps/RNNCell.cs

@@ -145,7 +145,7 @@ private Tensor _zero_state_tensors(object state_size, Tensor batch_size, TF_Data
             throw new NotImplementedException("_zero_state_tensors");
         }
 
-        public Tensors Apply(Tensors inputs, Tensors state = null, bool is_training = false, IOptionalArgs? optional_args = null)
+        public Tensors Apply(Tensors inputs, Tensors state = null, bool? is_training = false, IOptionalArgs? optional_args = null)
         {
             throw new NotImplementedException();
         }

src/TensorFlowNET.Keras/Engine/Layer.Apply.cs

@@ -13,7 +13,7 @@ public partial class Layer
         /// <param name="state"></param>
         /// <param name="training"></param>
         /// <returns></returns>
-        public virtual Tensors Apply(Tensors inputs, Tensors states = null, bool training = false, IOptionalArgs? optional_args = null)
+        public virtual Tensors Apply(Tensors inputs, Tensors states = null, bool? training = false, IOptionalArgs? optional_args = null)
         {
             if (callContext.Value == null)
                 callContext.Value = new CallContext();

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

@@ -142,6 +142,7 @@ public History fit(IDatasetV2 dataset,
             int verbose = 1,
             List<ICallback> callbacks = null,
             IDatasetV2 validation_data = null,
+            int validation_step = 10,   // 间隔多少次会进行一次验证
             bool shuffle = true,
             int initial_epoch = 0,
             int max_queue_size = 10,
@@ -164,11 +165,11 @@ public History fit(IDatasetV2 dataset,
             });
 
 
-            return FitInternal(data_handler, epochs, verbose, callbacks, validation_data: validation_data,
+            return FitInternal(data_handler, epochs, validation_step, verbose, callbacks, validation_data: validation_data,
                     train_step_func: train_step_function);
         }
 
-        History FitInternal(DataHandler data_handler, int epochs, int verbose, List<ICallback> callbackList, IDatasetV2 validation_data, 
+        History FitInternal(DataHandler data_handler, int epochs, int validation_step, int verbose, List<ICallback> callbackList, IDatasetV2 validation_data, 
             Func<DataHandler, OwnedIterator, Dictionary<string, float>> train_step_func)
         {
             stop_training = false;
@@ -207,6 +208,9 @@ History FitInternal(DataHandler data_handler, int epochs, int verbose, List<ICal
 
                 if (validation_data != null)
                 {
+                    if (validation_step > 0 && epoch ==0 || (epoch) % validation_step != 0)
+                        continue;
+
                     var val_logs = evaluate(validation_data);
                     foreach(var log in val_logs)
                     {

src/TensorFlowNET.Keras/Layers/Rnn/RNN.cs

@@ -393,7 +393,7 @@ protected override Tensors Call(Tensors inputs, Tensors initial_state = null, bo
             }
         }
 
-        public override Tensors Apply(Tensors inputs, Tensors initial_states = null, bool training = false, IOptionalArgs? optional_args = null)
+        public override Tensors Apply(Tensors inputs, Tensors initial_states = null, bool? training = false, IOptionalArgs? optional_args = null)
         {
             RnnOptionalArgs? rnn_optional_args = optional_args as RnnOptionalArgs;
             if (optional_args is not null && rnn_optional_args is null)

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

@@ -58,6 +58,7 @@ public IDatasetV2 image_dataset_from_directory(string directory,
             if (shuffle)
                 dataset = dataset.shuffle(batch_size * 8, seed: seed);
             dataset = dataset.batch(batch_size);
+            dataset.class_names = class_name_list;
             return dataset;
         }
 

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

@@ -6,6 +6,31 @@ namespace Tensorflow.Keras
 {
     public partial class Preprocessing
     {
+
+        /// <summary>
+        /// 图片路径转为数据处理用的dataset
+        /// </summary>
+        /// <param name="image_paths"></param>
+        /// <param name="image_size"></param>
+        /// <param name="num_channels"></param>
+        /// <param name="interpolation">
+        /// 用于调整大小的插值方法。支持`bilinear`、`nearest`、`bicubic`、`area`、`lanczos3`、`lanczos5`、`gaussian`、`mitchellcubic`。
+        /// 默认为`'bilinear'`。
+        /// </param>
+        /// <returns></returns>
+        public IDatasetV2 paths_to_dataset(string[] image_paths,
+                    Shape image_size,
+                    int num_channels = 3,
+                    int num_classes = 6,
+                    string interpolation = "bilinear")
+        {
+            var path_ds = tf.data.Dataset.from_tensor_slices(image_paths);
+            var img_ds = path_ds.map(x => path_to_image(x, image_size, num_channels, interpolation));
+            var label_ds = dataset_utils.labels_to_dataset(new int[num_classes] , "", num_classes);
+
+            return img_ds;
+        }
+
         public IDatasetV2 paths_and_labels_to_dataset(string[] image_paths,
             Shape image_size,
             int num_channels,