fix array_ops.squeeze

.github/FUNDING.yml

@@ -9,4 +9,4 @@ community_bridge: # Replace with a single Community Bridge project-name e.g., cl
 liberapay: # Replace with a single Liberapay username
 issuehunt: # Replace with a single IssueHunt username
 otechie: # Replace with a single Otechie username
-custom: ['https://bit.ly/2op1mu5']# Replace with up to 4 custom sponsorship URLs e.g., ['link1', 'link2']
+custom: []# Replace with up to 4 custom sponsorship URLs e.g., ['link1', 'link2']

src/TensorFlowNET.Core/Gradients/Tape.ComputeGradient.cs

@@ -26,7 +26,6 @@ public Tensor[] ComputeGradient(Tensor[] target_tensor_ids,
                 tensor_tape_,
                 state.op_tape);
 
-            int gcCollectFlag = 0;
             while (!op_stack.empty())
             {
                 var op = op_stack.Dequeue();
@@ -155,9 +154,6 @@ public Tensor[] ComputeGradient(Tensor[] target_tensor_ids,
                             op_stack.Enqueue(op_id);
                     }
                 }
-
-                if (gcCollectFlag++ % 10 == 0)
-                    GC.Collect();
             }
 
             if (state.op_tape.Count > 0)

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

@@ -90,19 +90,20 @@ public Tensor Apply(Tensors input, Tensor filters)
                     strides.Insert(0, 1);
                     dilations.Insert(0, 1);
 
-                    var expanded = tf.expand_dims(input, spatial_start_dim);
+                    input = array_ops.expand_dims(input, spatial_start_dim);
+                    filters = array_ops.expand_dims(filters, 0);
 
                     result = gen_nn_ops.conv2d(new Conv2dParams
                     {
-                        Input = expanded,
+                        Input = input,
                         Filter = filters,
                         Strides = strides.ToArray(),
                         Padding = padding,
                         DataFormat = channel_first ? "NCHW" : "NHWC",
                         Dilations = dilations.ToArray(),
                         Name = name
                     });
-                    result = tf.squeeze(result, squeeze_dims: spatial_start_dim);
+                    result = array_ops.squeeze(result, new[] { spatial_start_dim });
                 }
             });
 

src/TensorFlowNET.Core/Operations/array_ops.cs

@@ -730,7 +730,7 @@ public static Tensor strided_slice_grad(Tensor shape, Tensor begin, Tensor end,
         /// <returns>A `Tensor`. Has the same type as `input`.
         /// Contains the same data as `input`, but has one or more dimensions of
         /// size 1 removed.</returns>
-        public static Tensor squeeze(Tensor input, int[] axis = null, string name = null, int[] squeeze_dims = null)
+        public static Tensor squeeze(Tensor input, int[] axis = null, string name = null)
             => gen_array_ops.squeeze(input, axis, name);
 
         public static Tensor identity(Tensor input, string name = null)

src/TensorFlowNET.Keras/Layers/Convolution/Convolutional.cs

@@ -48,11 +48,11 @@ public class Convolutional : Layer
         public Convolutional(ConvolutionalArgs args) : base(args)
         {
             this.args = args;
-            args.KernelSize = conv_utils.normalize_tuple(args.KernelSize.dims.Select(x => (int)x).ToArray(), args.Rank, "kernel_size");
-            args.Strides = conv_utils.normalize_tuple(args.Strides.dims.Select(x => (int)x).ToArray(), args.Rank, "strides");
+            args.KernelSize = conv_utils.normalize_tuple(args.KernelSize.as_int_list(), args.Rank, "kernel_size");
+            args.Strides = conv_utils.normalize_tuple(args.Strides.as_int_list(), args.Rank, "strides");
             args.Padding = conv_utils.normalize_padding(args.Padding);
             args.DataFormat = conv_utils.normalize_data_format(args.DataFormat);
-            args.DilationRate = conv_utils.normalize_tuple(args.DilationRate.dims.Select(x => (int)x).ToArray(), args.Rank, "dilation_rate");
+            args.DilationRate = conv_utils.normalize_tuple(args.DilationRate.as_int_list(), args.Rank, "dilation_rate");
             inputSpec = new InputSpec(ndim: rank + 2);
             _tf_data_format = conv_utils.convert_data_format(data_format, rank + 2);
         }

src/TensorFlowNET.Keras/Layers/LayersApi.cs

@@ -68,64 +68,6 @@ public BatchNormalization BatchNormalization(int axis = -1,
                     Name = name
                 });
 
-        /// <summary>
-        /// 1D convolution layer (e.g. temporal convolution).
-        /// This layer creates a convolution kernel that is convolved with the layer input over a single spatial(or temporal) dimension to produce a tensor of outputs.If use_bias is True, a bias vector is created and added to the outputs.Finally, if activation is not None, it is applied to the outputs as well.
-        /// </summary>
-        /// <param name="filters">Integer, the dimensionality of the output space (i.e. the number of output filters in the convolution)</param>
-        /// <param name="kernel_size">An integer specifying the width of the 1D convolution window.</param>
-        /// <param name="strides">An integer specifying the stride of the convolution window . Specifying any stride value != 1 is incompatible with specifying any dilation_rate value != 1.</param>
-        /// <param name="padding">one of "valid" or "same" (case-insensitive). "valid" means no padding. "same" results in padding evenly to the left/right or up/down of the input such that output has the same height/width dimension as the input.</param>
-        /// <param name="data_format">A string, one of channels_last (default) or channels_first. The ordering of the dimensions in the inputs. channels_last corresponds to inputs with shape (batch_size, height, width, channels) while channels_first corresponds to inputs with shape (batch_size, channels, height, width). It defaults to the image_data_format value found in your Keras config file at ~/.keras/keras.json. If you never set it, then it will be channels_last.</param>
-        /// <param name="dilation_rate">An integer specifying the dilation rate to use for dilated convolution.Currently, specifying any dilation_rate value != 1 is incompatible with specifying any stride value != 1.</param>
-        /// <param name="groups">A positive integer specifying the number of groups in which the input is split along the channel axis. Each group is convolved separately with filters / groups filters. The output is the concatenation of all the groups results along the channel axis. Input channels and filters must both be divisible by groups.</param>
-        /// <param name="activation">Activation function to use. If you don't specify anything, no activation is applied (see keras.activations).</param>
-        /// <param name="use_bias">Boolean, whether the layer uses a bias vector.</param>
-        /// <param name="kernel_initializer">Initializer for the kernel weights matrix (see keras.initializers).</param>
-        /// <param name="bias_initializer">Initializer for the bias vector (see keras.initializers).</param>
-        /// <param name="kernel_regularizer">Regularizer function applied to the kernel weights matrix (see keras.regularizers).</param>
-        /// <param name="bias_regularizer">Regularizer function applied to the bias vector (see keras.regularizers).</param>
-        /// <param name="activity_regularizer">Regularizer function applied to the output of the layer (its "activation") (see keras.regularizers).</param>
-        /// <returns>A tensor of rank 3 representing activation(conv1d(inputs, kernel) + bias).</returns>
-        public Conv1D Conv1D(int filters,
-            int? kernel_size = null,
-            int? strides = null,
-            string padding = "valid",
-            string data_format = null,
-            int? dilation_rate = null,
-            int groups = 1,
-            Activation activation = null,
-            bool use_bias = true,
-            IInitializer kernel_initializer = null,
-            IInitializer bias_initializer = null,
-            IRegularizer kernel_regularizer = null,
-            IRegularizer bias_regularizer = null,
-            IRegularizer activity_regularizer = null)
-        {
-            // Special case: Conv1D will be implemented as Conv2D with H=1, so we need to add a 1-sized dimension to the kernel.
-            // Lower-level logic handles the stride and dilation_rate, but the kernel_size needs to be set properly here.
-
-            var kernel = (kernel_size == null) ? (1, 5) : (1, kernel_size.Value);
-            return new Conv1D(new Conv1DArgs
-            {
-                Rank = 1,
-                Filters = filters,
-                KernelSize = kernel,
-                Strides = strides == null ? 1 : strides,
-                Padding = padding,
-                DataFormat = data_format,
-                DilationRate = dilation_rate == null ? 1 : dilation_rate,
-                Groups = groups,
-                UseBias = use_bias,
-                KernelInitializer = kernel_initializer == null ? tf.glorot_uniform_initializer : kernel_initializer,
-                BiasInitializer = bias_initializer == null ? tf.zeros_initializer : bias_initializer,
-                KernelRegularizer = kernel_regularizer,
-                BiasRegularizer = bias_regularizer,
-                ActivityRegularizer = activity_regularizer,
-                Activation = activation ?? keras.activations.Linear
-            });
-        }
-
         /// <summary>
         /// 1D convolution layer (e.g. temporal convolution).
         /// This layer creates a convolution kernel that is convolved with the layer input over a single spatial(or temporal) dimension to produce a tensor of outputs.If use_bias is True, a bias vector is created and added to the outputs.Finally, if activation is not None, it is applied to the outputs as well.
@@ -143,7 +85,7 @@ public Conv1D Conv1D(int filters,
         /// <param name="bias_initializer">Initializer for the bias vector (see keras.initializers).</param>
         /// <returns>A tensor of rank 3 representing activation(conv1d(inputs, kernel) + bias).</returns>
         public Conv1D Conv1D(int filters,
-            int? kernel_size = null,
+            Shape? kernel_size = null,
             int? strides = null,
             string padding = "valid",
             string data_format = null,
@@ -157,12 +99,11 @@ public Conv1D Conv1D(int filters,
             // Special case: Conv1D will be implemented as Conv2D with H=1, so we need to add a 1-sized dimension to the kernel.
             // Lower-level logic handles the stride and dilation_rate, but the kernel_size needs to be set properly here.
 
-            var kernel = (kernel_size == null) ? (1, 5) : (1, kernel_size.Value);
             return new Conv1D(new Conv1DArgs
             {
                 Rank = 1,
                 Filters = filters,
-                KernelSize = kernel,
+                KernelSize = kernel_size ?? new Shape(1, 5),
                 Strides = strides == null ? 1 : strides,
                 Padding = padding,
                 DataFormat = data_format,

src/TensorFlowNET.Keras/Saving/hdf5_format.cs

@@ -210,19 +210,19 @@ public static void save_weights_to_hdf5_group(long f, List<ILayer> layers)
             }
         }
 
-        private static void save_attributes_to_hdf5_group(long f,string name ,Array data)
+        private static void save_attributes_to_hdf5_group(long f, string name, Array data)
         {
             int num_chunks = 1;
-           
+
             var chunked_data = Split(data, num_chunks);
-            int getSize= 0;
-           
-            string getType = data.Length>0?data.GetValue(0).GetType().Name.ToLower():"string";
+            int getSize = 0;
+
+            string getType = data.Length > 0 ? data.GetValue(0).GetType().Name.ToLower() : "string";
 
             switch (getType)
             {
                 case "single":
-                    getSize=sizeof(float);
+                    getSize = sizeof(float);
                     break;
                 case "double":
                     getSize = sizeof(double);
@@ -237,30 +237,25 @@ private static void save_attributes_to_hdf5_group(long f,string name ,Array data
                     getSize = sizeof(long);
                     break;
                 default:
-                    getSize=-1;
+                    getSize = -1;
                     break;
             }
             int getCount = chunked_data.Count;
-
-            if (getSize != -1) {
-                num_chunks = (int)Math.Ceiling((double)(getCount * getSize) / (double)HDF5_OBJECT_HEADER_LIMIT);
+
+            if (getSize != -1)
+            {
+                num_chunks = (int)Math.Ceiling((double)(getCount * getSize) / HDF5_OBJECT_HEADER_LIMIT);
                 if (num_chunks > 1) chunked_data = Split(data, num_chunks);
             }
-            
+
             if (num_chunks > 1)
             {
                 foreach (var (chunk_id, chunk_data) in enumerate(chunked_data))
-                {
-
                     WriteAttrs(f, getType, $"{name}{chunk_id}", chunk_data.ToArray());
-
-                }
-
             }
-            else {
-
-                WriteAttrs(f, getType,name, data);
-
+            else
+            {
+                WriteAttrs(f, getType, name, data);
             }
         }