Support of tensorflow.keras instead of keras

.gitignore

@@ -5,6 +5,7 @@ __pycache__/
 .idea/
 .DS_Store
 
+*.h5
 *.tsv
 *.tar.gz
 *out*

requirements.txt

@@ -1,4 +1,5 @@
 keras==2.3.1
-numpy==1.17.3
+numpy==1.16.2
 # tensorflow-gpu==1.14.0
 h5py
+gast==0.2.2

setup.py

@@ -2,14 +2,17 @@
 
 setup(
     name='keras-tcn',
-    version='2.8.3',
+    version='2.9.0',
     description='Keras TCN',
     author='Philippe Remy',
     license='MIT',
     long_description_content_type='text/markdown',
     long_description=open('README.md').read(),
     packages=['tcn'],
     # manually install tensorflow or tensorflow-gpu
-    install_requires=['numpy',
-                      'keras']
+    install_requires=[
+        'numpy==1.16.2',
+        'keras',
+        'gast==0.2.2'
+    ]
 )

tasks/imdb_tcn.py

@@ -5,10 +5,10 @@
 Based on: https://github.com/keras-team/keras/blob/master/examples/imdb_bidirectional_lstm.py
 """
 import numpy as np
-from keras import Model, Input
-from keras.datasets import imdb
-from keras.layers import Dense, Dropout, Embedding
-from keras.preprocessing import sequence
+from tensorflow.keras import Model, Input
+from tensorflow.keras.datasets import imdb
+from tensorflow.keras.layers import Dense, Dropout, Embedding
+from tensorflow.keras.preprocessing import sequence
 
 from tcn import TCN
 

tasks/save_reload_model.py

@@ -1,40 +1,42 @@
-import os
-
 import numpy as np
-from keras import Model, Input
-from keras.layers import Dense, Dropout, Embedding
-
-from tcn import TCN
+from tensorflow.keras.models import Sequential, model_from_json
+from tensorflow.keras.layers import Dense, Dropout, Embedding
+from tcn.tcn import TCN
 
-# simple TCN model.
+# define input shape
 max_len = 100
 max_features = 50
-i = Input(shape=(max_len,))
-x = Embedding(max_features, 16)(i)
-x = TCN(nb_filters=12,
-        dropout_rate=0.5,  # with dropout here.
-        kernel_size=6,
-        dilations=[1, 2, 4])(x)
-x = Dropout(0.5)(x)  # and dropout here.
-x = Dense(1, activation='sigmoid')(x)
 
-model = Model(inputs=[i], outputs=[x])
+# make model
+model = Sequential(layers=[Embedding(max_features, 16, input_shape=(max_len,)),
+                           TCN(nb_filters=12,
+                               dropout_rate=0.5,
+                               kernel_size=6,
+                               dilations=[1, 2, 4]),
+                           Dropout(0.5),
+                           Dense(units=1, activation='sigmoid')])
 
-if os.path.exists('tcn.npz'):
-    # Load checkpoint if file exists.
-    w = np.load('tcn.npz', allow_pickle=True)['w']
-    print('Model reloaded.')
-    model.set_weights(w.tolist())
-else:
-    # Save the checkpoint.
-    w = np.array(model.get_weights())
-    np.savez_compressed(file='tcn.npz', w=w, allow_pickle=True)
-    print('First time.')
+# get model as json string and save to file
+model_as_json = model.to_json()
+with open(r'model.json', "w") as json_file:
+    json_file.write(model_as_json)
+# save weights to file (for this format, need h5py installed)
+model.save_weights('weights.h5')
 
 # Make inference.
-# The value for [First time] and [Model reloaded] should be the same. Run the script twice!
 inputs = np.ones(shape=(1, 100))
 out1 = model.predict(inputs)[0, 0]
 print('*' * 80)
-print(out1)
+print('Inference after creation:', out1)
+
+# load model from file
+loaded_json = open(r'model.json', 'r').read()
+reloaded_model = model_from_json(loaded_json, custom_objects={'TCN': TCN})
+
+# restore weights
+reloaded_model.load_weights(r'weights.h5')
+
+# Make inference.
+out2 = reloaded_model.predict(inputs)[0, 0]
 print('*' * 80)
+print('Inference after loading:', out2)

tasks/save_reload_sequential_model.py

@@ -1,7 +1,7 @@
 import numpy as np
 from keras.models import Sequential, model_from_json
 from keras.layers import Dense, Dropout, Embedding
-from tcn.tcn import TCN
+from tcn import TCN
 
 # define input shape
 max_len = 100

tasks/sequential.py

@@ -5,11 +5,11 @@
 Based on: https://github.com/keras-team/keras/blob/master/examples/imdb_bidirectional_lstm.py
 """
 import numpy as np
-from keras import Sequential
-from keras.callbacks import Callback
-from keras.datasets import imdb
-from keras.layers import Dense, Dropout, Embedding
-from keras.preprocessing import sequence
+from tensorflow.keras import Sequential
+from tensorflow.keras.callbacks import Callback
+from tensorflow.keras.datasets import imdb
+from tensorflow.keras.layers import Dense, Dropout, Embedding
+from tensorflow.keras.preprocessing import sequence
 
 from tcn import TCN
 
@@ -50,7 +50,8 @@ class TestCallback(Callback):
 
     def on_epoch_end(self, epoch, logs=None):
         print(logs)
-        assert logs['val_accuracy'] > 0.78
+        acc_key = 'val_accuracy' if 'val_accuracy' in logs else 'val_acc'
+        assert logs[acc_key] > 0.78
 
 
 print('Train...')

tasks/time_series_forecasting.py

@@ -2,8 +2,8 @@
 import matplotlib.pyplot as plt
 import numpy as np
 import pandas as pd
-from keras import Input, Model
-from keras.layers import Dense
+from tensorflow.keras import Input, Model
+from tensorflow.keras.layers import Dense
 
 from tcn import TCN
 

tcn/tcn.py

@@ -1,13 +1,12 @@
 from typing import List
 
-import keras.backend as K
-import keras.layers
-from keras import optimizers
-from keras.layers import Layer
-from keras.layers import Activation, Lambda, add
-from keras.layers import Conv1D, SpatialDropout1D
-from keras.layers import Dense, BatchNormalization
-from keras.models import Input, Model
+from tensorflow.keras import backend as K, Model, Input, optimizers
+from tensorflow.keras import layers
+from tensorflow.keras.layers import Activation, SpatialDropout1D, Lambda
+from tensorflow.keras.layers import Layer
+from tensorflow.python.keras.layers import Conv1D
+from tensorflow.python.layers.core import Dense
+from tensorflow.python.layers.normalization import BatchNormalization
 
 
 class ResidualBlock(Layer):
@@ -58,7 +57,7 @@ def _add_and_activate_layer(self, layer):
 
         Args:
             layer: Appends layer to internal layer list and builds it based on the current output
-                   shape of ResidualBlock. Updates current output shape.
+                   shape of ResidualBlocK. Updates current output shape.
 
         """
         self.residual_layers.append(layer)
@@ -90,7 +89,7 @@ def build(self, input_shape):
 
             if not self.last_block:
                 # 1x1 conv to match the shapes (channel dimension).
-                name = 'conv1D_{}'.format(k+1)
+                name = 'conv1D_{}'.format(k + 1)
                 with K.name_scope(name):
                     # make and build this layer separately because it directly uses input_shape
                     self.shape_match_conv = Conv1D(filters=self.nb_filters,
@@ -100,7 +99,7 @@ def build(self, input_shape):
                                                    kernel_initializer=self.kernel_initializer)
 
             else:
-                 self.shape_match_conv = Lambda(lambda x: x, name='identity')
+                self.shape_match_conv = Lambda(lambda x: x, name='identity')
 
             self.shape_match_conv.build(input_shape)
             self.res_output_shape = self.shape_match_conv.compute_output_shape(input_shape)
@@ -128,7 +127,7 @@ def call(self, inputs, training=None):
                 x = layer(x)
 
         x2 = self.shape_match_conv(inputs)
-        res_x = add([x2, x])
+        res_x = layers.add([x2, x])
         return [self.final_activation(res_x), x]
 
     def compute_output_shape(self, input_shape):
@@ -159,7 +158,7 @@ class TCN(Layer):
             dilations: The list of the dilations. Example is: [1, 2, 4, 8, 16, 32, 64].
             nb_stacks : The number of stacks of residual blocks to use.
             padding: The padding to use in the convolutional layers, 'causal' or 'same'.
-            use_skip_connections: Boolean. If we want to add skip connections from input to each residual block.
+            use_skip_connections: Boolean. If we want to add skip connections from input to each residual blocK.
             return_sequences: Boolean. Whether to return the last output in the output sequence, or the full sequence.
             activation: The activation used in the residual blocks o = Activation(x + F(x)).
             dropout_rate: Float between 0 and 1. Fraction of the input units to drop.
@@ -237,7 +236,7 @@ def build(self, input_shape):
                                                           dropout_rate=self.dropout_rate,
                                                           use_batch_norm=self.use_batch_norm,
                                                           kernel_initializer=self.kernel_initializer,
-                                                          last_block = len(self.residual_blocks)+1 == total_num_blocks,
+                                                          last_block=len(self.residual_blocks) + 1 == total_num_blocks,
                                                           name='residual_block_{}'.format(len(self.residual_blocks))))
                 # build newest residual block
                 self.residual_blocks[-1].build(self.build_output_shape)
@@ -270,7 +269,7 @@ def call(self, inputs, training=None):
             skip_connections.append(skip_out)
 
         if self.use_skip_connections:
-            x = add(skip_connections)
+            x = layers.add(skip_connections)
         if not self.return_sequences:
             x = self.lambda_layer(x)
         return x
@@ -303,7 +302,7 @@ def compiled_tcn(num_feat,  # type: int
                  dilations,  # type: List[int]
                  nb_stacks,  # type: int
                  max_len,  # type: int
-                 output_len=1, #type: int
+                 output_len=1,  # type: int
                  padding='causal',  # type: str
                  use_skip_connections=True,  # type: bool
                  return_sequences=True,
@@ -328,7 +327,7 @@ def compiled_tcn(num_feat,  # type: int
         nb_stacks : The number of stacks of residual blocks to use.
         max_len: The maximum sequence length, use None if the sequence length is dynamic.
         padding: The padding to use in the convolutional layers.
-        use_skip_connections: Boolean. If we want to add skip connections from input to each residual block.
+        use_skip_connections: Boolean. If we want to add skip connections from input to each residual blocK.
         return_sequences: Boolean. Whether to return the last output in the output sequence, or the full sequence.
         regression: Whether the output should be continuous or discrete.
         dropout_rate: Float between 0 and 1. Fraction of the input units to drop.