def create_model():
units = 512
middle_units = 256
dropout_value = 0.3
activation_function = 'softmax'
loss_function = 'categorical_crossentropy'
optimizer = 'rmsprop'
model = Sequential()
model.add(
LSTM(units,
input_shape=(network_input.shape[1], network_input.shape[2]),
recurrent_dropout=dropout_value,
return_sequences=True))
model.add(
LSTM(
units,
return_sequences=True,
recurrent_dropout=dropout_value,
))
model.add(LSTM(units))
model.add(BatchNormalization())
model.add(Dropout(dropout_value))
model.add(Dense(middle_units))
model.add(Activation('relu'))
model.add(Dropout(dropout_value))
model.add(BatchNormalization())
model.add(Dropout(dropout_value))
model.add(Dense(vocab_size))
model.add(Activation(activation_function))
model.compile(loss=loss_function, optimizer=optimizer)
return model
def modelStandard(input_shape, parameter=None):
# define LSTM
model = Sequential()
model.add(
TimeDistributed(Conv2D(16, (2, 2), activation='relu'),
input_shape=input_shape))
model.add(Dropout(parameter['dropout']))
model.add(BatchNormalization())
model.add(TimeDistributed(MaxPooling2D(pool_size=(2, 2), strides=2)))
model.add(Dropout(parameter['dropout']))
model.add(TimeDistributed(Flatten()))
model.add(LSTM(parameter['cell1']))
# model.add(Dropout(0.25))
model.add(BatchNormalization())
model.add(RepeatVector(8))
model.add(LSTM(parameter['cell2'], return_sequences=True))
# model.add(Dropout(0.25))
model.add(BatchNormalization())
model.add(TimeDistributed(Dense(5, activation='softmax')))
# Replicates `model` on 8 GPUs.
# This assumes that your machine has 8 available GPUs.
#parallel_model = multi_gpu_model(model, gpus=2)
#parallel_model.compile(loss='categorical_crossentropy',
# optimizer='adam', metrics=['accuracy'])
model.compile(loss='categorical_crossentropy',
optimizer='adam',
metrics=['accuracy'])
return model
def modelC(row, col):
# define LSTM
model = Sequential()
model.add(
TimeDistributed(Conv2D(16, (2, 2), activation='relu'),
input_shape=(None, row, col, 1)))
model.add(Dropout(0.25))
model.add(BatchNormalization())
model.add(TimeDistributed(MaxPooling2D(pool_size=(2, 2))))
model.add(Dropout(0.25))
model.add(TimeDistributed(Flatten()))
model.add(LSTM(75))
# model.add(Dropout(0.25))
model.add(BatchNormalization())
model.add(RepeatVector(4))
model.add(LSTM(50, return_sequences=True))
# model.add(Dropout(0.25))
model.add(BatchNormalization())
model.add(TimeDistributed(Dense(4, activation='softmax')))
# Replicates `model` on 8 GPUs.
# This assumes that your machine has 8 available GPUs.
# parallel_model = multi_gpu_model(model, gpus=[2])
# parallel_model.compile(loss='categorical_crossentropy',
# optimizer='adam', metrics=['accuracy'])
model.compile(loss='categorical_crossentropy',
optimizer='adam',
metrics=['accuracy'])
return model
def modelB(row, col, parameter=None):
# define LSTM
input = Input(shape=(None, row, col, 1), name='main_input')
''' x = TimeDistributed(Conv2D(16, (2, 2)))(input)
x = BatchNormalization()(x)
x = Activation('relu')(x)
x = Dropout(0.25)(x)
'''
# tower_1 = TimeDistributed(Conv2D(16, (1, 1), padding='same', activation='relu'))(input)
# tower_1 = TimeDistributed(Conv2D(16, (3, 3), padding='same', activation='relu'))(tower_1)
tower_2 = TimeDistributed(Conv2D(16, (1, 1), padding='same'))(input)
x = BatchNormalization()(tower_2)
x = Activation('relu')(x)
x = Dropout(0.25)(x)
tower_2 = TimeDistributed(Conv2D(16, (5, 5), padding='same'))(x)
x = BatchNormalization()(tower_2)
x = Activation('relu')(x)
tower_2 = Dropout(0.25)(x)
tower_3 = TimeDistributed(
MaxPooling2D((3, 3), strides=(1, 1), padding='same'))(input)
tower_3 = TimeDistributed(Conv2D(16, (1, 1), padding='same'))(tower_3)
x = BatchNormalization()(tower_3)
x = Activation('relu')(x)
tower_3 = Dropout(0.25)(x)
concatenate_output = concatenate([tower_2, tower_3], axis=-1)
x = TimeDistributed(MaxPooling2D(pool_size=(2, 2),
strides=2))(concatenate_output)
x = Dropout(0.25)(x)
x = TimeDistributed(Flatten())(x)
# convLstm = ConvLSTM2D(filters=40, kernel_size=(3, 3),padding='same', return_sequences=False)(x)
lstm_output = LSTM(75)(x)
lstm_output = BatchNormalization()(lstm_output)
# lstm_output = BatchNormalization()(convLstm)
# auxiliary_output = Dense(1, activation='sigmoid', name='aux_output')(lstm_output)
# auxiliary_input = Input(shape=(4,), name='aux_input')
# x = concatenate([lstm_output, auxiliary_input])
x = RepeatVector(4)(lstm_output)
x = LSTM(50, return_sequences=True)(x)
# model.add(Dropout(0.25))
x = BatchNormalization()(x)
output = TimeDistributed(Dense(4, activation='softmax'),
name='main_output')(x)
model = Model(inputs=[input], outputs=[output])
model.compile(loss={'main_output': 'categorical_crossentropy'},
loss_weights={'main_output': 1.},
optimizer='adam',
metrics=['accuracy'])
return model
def ModelShare():
tweet_a = Input(shape=(280, 256))
tweet_b = Input(shape=(280, 256))
# This layer can take as input a matrix
# and will return a vector of size 64
shared_lstm = LSTM(64, return_sequences=True, name='lstm')
# When we reuse the same layer instance
# multiple times, the weights of the layer
# are also being reused
# (it is effectively *the same* layer)
encoded_a = shared_lstm(tweet_a)
encoded_b = shared_lstm(tweet_b)
# We can then concatenate the two vectors:
merged_vector = concatenate([encoded_a, encoded_b], axis=-1)
# And add a logistic regression on top
predictions = Dense(1, activation='sigmoid')(merged_vector)
# We define a trainable model linking the
# tweet inputs to the predictions
model = Model(inputs=[tweet_a, tweet_b], outputs=predictions)
model.compile(optimizer='rmsprop',
loss='binary_crossentropy',
metrics=['accuracy'])
return model
def modelDemoStandardConvLSTMInception(input_shape, parameter=None):
# define LSTM
input = Input(shape=input_shape, name='main_input')
I_1 = TimeDistributed(Conv2D(16, (1, 1),
activation='relu',
padding='same',
name='C_1'),
name='I_11')(input)
I_1 = TimeDistributed(Conv2D(16, (5, 5),
activation='relu',
padding='same',
name='C_2'),
name='I_12')(I_1)
I_2 = TimeDistributed(MaxPooling2D((3, 3),
strides=(1, 1),
padding='same',
name='C_3'),
name='I_21')(input)
I_2 = TimeDistributed(Conv2D(16, (1, 1),
activation='relu',
padding='same',
name='C_4'),
name='I_22')(I_2)
concatenate_output = concatenate([I_1, I_2], axis=-1)
# x = TimeDistributed(Flatten())(x)
x = ConvLSTM2D(filters=32,
kernel_size=(3, 3),
padding='same',
return_sequences=False)(concatenate_output)
#x = MaxPooling2D((3, 3), strides=(1, 1), padding='same', name='M_1')(x)
x = (Flatten())(x)
x = RepeatVector(8)(x)
x = LSTM(50, return_sequences=True)(x)
output = TimeDistributed(Dense(8, activation='softmax'),
name='main_output')(x)
#with tensorflow.device('/cpu'):
model = Model(inputs=[input], outputs=[output])
# compile the model with gpu
#parallel_model = multi_gpu_model(model, gpus=2)
#parallel_model.compile(loss={'main_output': 'categorical_crossentropy'},
# loss_weights={'main_output': 1.}, optimizer='adam', metrics=['accuracy'])
#model = multi_gpu(model, gpus=[1, 2])
model.compile(loss={'main_output': 'categorical_crossentropy'},
loss_weights={'main_output': 1.},
optimizer='adam',
metrics=['accuracy'])
return model
def modelDemoStandard(row, col):
# define LSTM
input = Input(shape=(None, row, col, 1), name='main_input')
x = TimeDistributed(Conv2D(16, (2, 2), activation='relu'))(input)
x = TimeDistributed(Flatten())(x)
lstm_output = LSTM(75)(x)
x = RepeatVector(4)(lstm_output)
x = LSTM(50, return_sequences=True)(x)
output = TimeDistributed(Dense(4, activation='softmax'),
name='main_output')(x)
model = Model(inputs=[input], outputs=[output])
model.compile(loss={'main_output': 'categorical_crossentropy'},
loss_weights={'main_output': 1.},
optimizer='adam',
metrics=['accuracy'])
return model
def create_model():
# # ENCODER DECODER
# encoder_inputs = Input(shape=(128, 3))
# # masked_encoder_inputs = Masking()(encoder_inputs)
# masked_encoder_inputs = encoder_inputs
# encoder_lstm = Bidirectional(LSTM(500, return_state=True))
#
# # We discard `encoder_outputs` and only keep the states.
# _, forward_h, forward_c, backward_h, backward_c = encoder_lstm(masked_encoder_inputs)
# state_c = Concatenate()([forward_c, backward_c])
# state_h = Concatenate()([forward_h, backward_h])
# encoder_states = [state_h, state_c]
#
# # Bottleneck Here
#
# decoder_inputs = Input(shape=(128, 3))
# # masked_decoder_inputs = Masking()(decoder_inputs)
# masked_decoder_inputs = decoder_inputs
# decoder_lstm = LSTM(1000, return_state=True, return_sequences=True)
# decoder_outputs, _, _ = decoder_lstm(masked_decoder_inputs, initial_state=encoder_states)
#
# outputs = TimeDistributed(Dense(2, activation=capped_relu))(decoder_outputs)
#
# model = Model([encoder_inputs, decoder_inputs], outputs)
# SEQUENTIAL
model = models.Sequential([
Input((128, 3)),
Bidirectional(LSTM(500, return_sequences=True)),
Bidirectional(LSTM(500, return_sequences=True)),
TimeDistributed(Dense(2, activation=capped_relu))
])
model.compile(optimizer=optimizers.Adam(learning_rate=0.0001),
loss=losses.MeanAbsoluteError(),
metrics=["accuracy"])
print(model.summary())
return model
def modelA(row, col):
# define LSTM
input = Input(shape=(None, row, col, 1), name='main_input')
x = TimeDistributed(Conv2D(16, (2, 2), activation='relu'))(input)
x = Dropout(0.25)(x)
x = BatchNormalization()(x)
x = TimeDistributed(MaxPooling2D(pool_size=(2, 2), strides=2))(x)
x = Dropout(0.25)(x)
x = TimeDistributed(Flatten())(x)
lstm_output = LSTM(75)(x)
lstm_output = BatchNormalization()(lstm_output)
auxiliary_output = Dense(1, activation='sigmoid',
name='aux_output')(lstm_output)
auxiliary_input = Input(shape=(4, ), name='aux_input')
x = concatenate([lstm_output, auxiliary_input])
x = RepeatVector(8)(x)
x = LSTM(50, return_sequences=True)(x)
# model.add(Dropout(0.25))
x = BatchNormalization()(x)
output = TimeDistributed(Dense(5, activation='softmax'),
name='main_output')(x)
model = Model(inputs=[input, auxiliary_input],
outputs=[output, auxiliary_output])
model.compile(loss={
'main_output': 'categorical_crossentropy',
'aux_output': 'binary_crossentropy'
},
loss_weights={
'main_output': 1.,
'aux_output': 0.2
},
optimizer='adam',
metrics=['accuracy'])
return model
def modelStandardB(row, col):
# define LSTM
input_img = Input(shape=(None, row, col, 1), name='input')
x = TimeDistributed(Conv2D(16, (2, 2), activation='relu'))(input_img)
x = Dropout(0.25)(x)
x = BatchNormalization()(x)
x = TimeDistributed(MaxPooling2D(pool_size=(2, 2), strides=2))(x)
x = Dropout(0.25)(x)
x = TimeDistributed(Flatten())(x)
x = LSTM(75)(x)
# model.add(Dropout(0.25))
x = BatchNormalization()(x)
x = RepeatVector(4)(x)
x = LSTM(50, return_sequences=True)(x)
# model.add(Dropout(0.25))
x = BatchNormalization()(x)
output = TimeDistributed(Dense(4, activation='softmax'))(x)
model = Model(input_img, output)
model.compile(loss='categorical_crossentropy',
optimizer='adam',
metrics=['accuracy'])
return model
def modelMasking(self, code_layer_type, input_dim, code_dim):
self.code_layer_type = code_layer_type
assert len(code_dim) > 0
if self.code_layer_type == 'lstm':
assert len(input_dim) == 2
input_data = Input(shape=(input_dim[0], input_dim[1]))
mask = Masking(mask_value=0.)(input_data)
if len(code_dim) == 1:
encoded = LSTM(code_dim[0])(mask)
decoded = RepeatVector(input_dim[0])(encoded)
elif len(code_dim) > 1:
encoded = mask
for i, units in enumerate(code_dim):
if i == len(code_dim) - 1:
encoded = LSTM(units)(encoded)
continue
encoded = LSTM(units, return_sequences=True)(encoded)
for i, units in enumerate(reversed(code_dim)):
if i == 1:
decoded = LSTM(units, return_sequences=True)(
RepeatVector(input_dim[0])(encoded))
elif i > 1:
decoded = LSTM(units, return_sequences=True)(decoded)
else:
raise ValueError("The codDim must be over 0.")
decoded = LSTM(input_dim[-1], return_sequences=True)(decoded)
self.model = Model(input_data, decoded)
elif self.code_layer_type == 'cov':
pass
elif self.code_layer_type == 'dense':
assert len(input_dim) == 1
input_data = Input(shape=(input_dim[0], ))
# encoded = input_data
# for i, units in enumerate(codeDim):
# encoded = Dense(units, activation='relu')(encoded)
# decoded = Dense(inputDim[-1], activation='sigmoid')(encoded)
# self.model = Model(input_data, decoded)
encoder = Dense(
code_dim[0],
activation="tanh",
activity_regularizer=regularizers.l1(10e-5))(input_data)
encoder = Dense(int(code_dim[0] / 2), activation="relu")(encoder)
decoder = Dense(int(code_dim[0] / 2), activation='tanh')(encoder)
decoder = Dense(input_dim[0], activation='relu')(decoder)
self.model = Model(input_data, decoder)
def ModelVisualQuestionAnswering():
# First, let's define a vision model using a Sequential model.
# This model will encode an image into a vector.
vision_model = Sequential()
vision_model.add(
Conv2D(64, (3, 3),
activation='relu',
padding='same',
input_shape=(224, 224, 3)))
vision_model.add(Conv2D(64, (3, 3), activation='relu'))
vision_model.add(MaxPooling2D((2, 2)))
vision_model.add(Conv2D(128, (3, 3), activation='relu', padding='same'))
vision_model.add(Conv2D(128, (3, 3), activation='relu'))
vision_model.add(MaxPooling2D((2, 2)))
vision_model.add(Conv2D(256, (3, 3), activation='relu', padding='same'))
vision_model.add(Conv2D(256, (3, 3), activation='relu'))
vision_model.add(Conv2D(256, (3, 3), activation='relu'))
vision_model.add(MaxPooling2D((2, 2)))
vision_model.add(Flatten())
# Now let's get a tensor with the output of our vision model:
image_input = Input(shape=(224, 224, 3))
encoded_image = vision_model(image_input)
# Next, let's define a language model to encode the question into a vector.
# Each question will be at most 100 words long,
# and we will index words as integers from 1 to 9999.
question_input = Input(shape=(100, ), dtype='int32')
embedded_question = Embedding(input_dim=10000,
output_dim=256,
input_length=100)(question_input)
encoded_question = LSTM(256)(embedded_question)
# Let's concatenate the question vector and the image vector:
merged = concatenate([encoded_question, encoded_image])
# And let's train a logistic regression over 1000 words on top:
output = Dense(1000, activation='softmax')(merged)
# This is our final model:
vqa_model = Model(inputs=[image_input, question_input], outputs=output)
return vqa_model
def get_model(X, N_class, total_words=86627, EMBEDDING_DIM=100, maxlen=53):
embeddings_index = {}
f = open('glove.6B/glove.6B.100d.txt')
for line in f:
values = line.split()
word = values[0]
coefs = np.asarray(values[1:], dtype='float32')
embeddings_index[word] = coefs
f.close()
embedding_matrix = np.zeros((total_words, EMBEDDING_DIM))
for word, i in X.items():
embedding_vector = embeddings_index.get(word)
if embedding_vector is not None:
embedding_matrix[i] = embedding_vector
inp = Input(shape=(maxlen, ), dtype='int32')
embedding = Embedding(total_words,
EMBEDDING_DIM,
embeddings_initializer=Constant(embedding_matrix),
input_length=maxlen,
trainable=False)(inp)
x = LSTM(300, dropout=0.25, recurrent_dropout=0.25,
return_sequences=True)(embedding)
x = Dropout(0.25)(x)
merged = Attention_COSTUM(maxlen)(x)
merged = Dense(256, activation='relu')(merged)
merged = Dropout(0.25)(merged)
merged = BatchNormalization()(merged)
outp = Dense(N_class, activation='softmax')(merged)
AttentionLSTM = Model(inputs=inp, outputs=outp)
AttentionLSTM.compile(loss='sparse_categorical_crossentropy',
optimizer='adam',
metrics=['acc'])
return AttentionLSTM
def modelDemoStandardConvLSTM(row, col):
# define LSTM
input = Input(shape=(None, row, col, 1), name='main_input')
# x = TimeDistributed(Flatten())(x)
x = ConvLSTM2D(filters=75,
kernel_size=(3, 3),
padding='same',
return_sequences=False)(input)
x = (Flatten())(x)
x = RepeatVector(4)(x)
x = LSTM(50, return_sequences=True)(x)
output = TimeDistributed(Dense(4, activation='softmax'),
name='main_output')(x)
model = Model(inputs=[input], outputs=[output])
model.compile(loss={'main_output': 'categorical_crossentropy'},
loss_weights={'main_output': 1.},
optimizer='adam',
metrics=['accuracy'])
return model
def model(self, code_layer_type, input_dim, code_dim):
self.code_layer_type = code_layer_type
assert len(code_dim) > 0
if self.code_layer_type == 'lstm':
assert len(input_dim) == 2
input_data = Input(shape=(input_dim[0], input_dim[1]))
if len(code_dim) == 1:
encoded = LSTM(code_dim[0])(input_data)
decoded = RepeatVector(input_dim[0])(encoded)
elif len(code_dim) > 1:
encoded = input_data
for i, units in enumerate(code_dim):
if i == len(code_dim) - 1:
encoded = LSTM(units)(encoded)
continue
encoded = LSTM(units, return_sequences=True)(encoded)
for i, units in enumerate(reversed(code_dim)):
if i == 1:
decoded = LSTM(units, return_sequences=True)(
RepeatVector(input_dim[0])(encoded))
elif i > 1:
decoded = LSTM(units, return_sequences=True)(decoded)
else:
raise ValueError("The codDim must be over 0.")
decoded = LSTM(input_dim[-1], return_sequences=True)(decoded)
self.model = Model(input_data, decoded)
elif self.code_layer_type == 'dense':
assert len(input_dim) == 1
input_data = Input(shape=(input_dim[0], ))
encoded = input_data
for i, units in enumerate(code_dim):
encoded = Dense(units, activation='relu')(encoded)
decoded = Dense(input_dim[-1], activation='sigmoid')(encoded)
self.model = Model(input_data, decoded)
elif self.code_layer_type == 'cov':
pass
def ModelVideoQuestionAnswering():
# First, let's define a vision model using a Sequential model.
# This model will encode an image into a vector.
vision_model = Sequential()
vision_model.add(
Conv2D(64, (3, 3),
activation='relu',
padding='same',
input_shape=(224, 224, 3)))
vision_model.add(Conv2D(64, (3, 3), activation='relu'))
vision_model.add(MaxPooling2D((2, 2)))
vision_model.add(Conv2D(128, (3, 3), activation='relu', padding='same'))
vision_model.add(Conv2D(128, (3, 3), activation='relu'))
vision_model.add(MaxPooling2D((2, 2)))
vision_model.add(Conv2D(256, (3, 3), activation='relu', padding='same'))
vision_model.add(Conv2D(256, (3, 3), activation='relu'))
vision_model.add(Conv2D(256, (3, 3), activation='relu'))
vision_model.add(MaxPooling2D((2, 2)))
vision_model.add(Flatten())
# Now let's get a tensor with the output of our vision model:
image_input = Input(shape=(224, 224, 3))
encoded_image = vision_model(image_input)
# Next, let's define a language model to encode the question into a vector.
# Each question will be at most 100 words long,
# and we will index words as integers from 1 to 9999.
question_input = Input(shape=(100, ), dtype='int32')
embedded_question = Embedding(input_dim=10000,
output_dim=256,
input_length=100)(question_input)
encoded_question = LSTM(256)(embedded_question)
# Let's concatenate the question vector and the image vector:
merged = concatenate([encoded_question, encoded_image])
# And let's train a logistic regression over 1000 words on top:
output = Dense(1000, activation='softmax')(merged)
# This is our final model:
# vqa_model = Model(inputs=[image_input, question_input], outputs=output)
video_input = Input(shape=(100, 224, 224, 3))
# This is our video encoded via the previously trained vision_model (weights are reused)
encoded_frame_sequence = TimeDistributed(vision_model)(
video_input) # the output will be a sequence of vectors
encoded_video = LSTM(256)(
encoded_frame_sequence) # the output will be a vector
# This is a model-level representation of the question encoder, reusing the same weights as before:
question_encoder = Model(inputs=question_input, outputs=encoded_question)
# Let's use it to encode the question:
video_question_input = Input(shape=(100, ), dtype='int32')
encoded_video_question = question_encoder(video_question_input)
# And this is our video question answering model:
merged = concatenate([encoded_video, encoded_video_question])
output = Dense(1000, activation='softmax')(merged)
video_qa_model = Model(inputs=[video_input, video_question_input],
outputs=output)
return video_qa_model
def build_model(self, model_name, query_dim, terms_dim, output_dim,
word_embedding):
self.model_name = model_name
query_input = Input(shape=(query_dim, ), name='query_input')
terms_input = Input(shape=(terms_dim, ), name='terms_input')
if model_name == 'lstm':
embedding_feature_block = Sequential(layers=[
Embedding(word_embedding.vocabulary_size,
word_embedding.dimensions,
weights=[word_embedding.embedding_matrix],
trainable=True,
mask_zero=False),
BatchNormalization(),
LSTM(64, return_sequences=True)
])
elif model_name == 'bilstm':
embedding_feature_block = Sequential(layers=[
Embedding(word_embedding.vocabulary_size,
word_embedding.dimensions,
weights=[word_embedding.embedding_matrix],
trainable=True,
mask_zero=False),
BatchNormalization(),
Bidirectional(LSTM(64, return_sequences=True))
])
else: # default cnn
embedding_feature_block = Sequential(layers=[
Embedding(word_embedding.vocabulary_size,
word_embedding.dimensions,
weights=[word_embedding.embedding_matrix],
trainable=True,
mask_zero=False),
BatchNormalization(),
Conv1D(filters=64, kernel_size=3, strides=1),
MaxPooling1D(pool_size=3)
])
# Features
query_feature = embedding_feature_block(query_input)
terms_feature = embedding_feature_block(terms_input)
# Query-Terms alignment
attention = Dot(axes=-1)([query_feature, terms_feature])
softmax_attention = Lambda(lambda x: softmax(x, axis=1),
output_shape=unchanged_shape)(attention)
terms_aligned = Dot(axes=1)([softmax_attention, terms_feature])
# Aligned features
if model_name == 'lstm':
flatten_layer = LSTM(128, return_sequences=False)(terms_aligned)
elif model_name == 'bilstm':
flatten_layer = Bidirectional(LSTM(
128, return_sequences=False))(terms_aligned)
else: # default cnn
merged_cnn = Conv1D(filters=128, kernel_size=3,
strides=1)(terms_aligned)
merged_cnn = MaxPooling1D(pool_size=3)(merged_cnn)
flatten_layer = Flatten()(merged_cnn)
# Output
dense = BatchNormalization()(flatten_layer)
dense = Dense(64, activation='sigmoid')(dense)
out = Dense(output_dim, activation='linear')(dense)
self.model = Model(inputs=[query_input, terms_input], outputs=out)
self.model.compile(optimizer='adam', loss=losses.mean_squared_error)
self.model.summary()
def construct_keras_api_model(embedding_weights):
# input_no_time_no_repeat = Input(shape=max_len, dtype='int32')
# embedded_no_time_no_repeat = Embedding(
# creative_id_window,embedding_size,weights=[embedding_weights],trainable=False
# )(input_no_time_no_repeat)
# ==================================================================================
Input_fix_creative_id = Input(shape=(math.ceil(time_id_max / period_days) *
period_length),
dtype='int32',
name='input_fix_creative_id')
Embedded_fix_creative_id = Embedding(
creative_id_window,
embedding_size,
weights=[embedding_weights],
trainable=False)(Input_fix_creative_id)
# ==================================================================================
# input_no_time_with_repeat = Input(shape=max_len, dtype='int32')
# embedded_no_time_with_repeat = Embedding(creative_id_window,embedding_size,weights=[embedding_weights],trainable=False)(input_no_time_with_repeat)
# ----------------------------------------------------------------------
GM_x = keras.layers.GlobalMaxPooling1D()(Embedded_fix_creative_id)
GM_x = Dropout(0.5)(GM_x)
GM_x = Dense(embedding_size // 2, kernel_regularizer=l2(0.001))(GM_x)
GM_x = BatchNormalization()(GM_x)
GM_x = Activation('relu')(GM_x)
GM_x = Dropout(0.5)(GM_x)
GM_x = Dense(embedding_size // 4, kernel_regularizer=l2(0.001))(GM_x)
GM_x = BatchNormalization()(GM_x)
GM_x = Activation('relu')(GM_x)
GM_x = Dense(1, 'sigmoid')(GM_x)
# ----------------------------------------------------------------------
GA_x = GlobalAveragePooling1D()(Embedded_fix_creative_id)
GA_x = Dropout(0.5)(GA_x)
GA_x = Dense(embedding_size // 2, kernel_regularizer=l2(0.001))(GA_x)
GA_x = BatchNormalization()(GA_x)
GA_x = Activation('relu')(GA_x)
GA_x = Dropout(0.5)(GA_x)
GA_x = Dense(embedding_size // 4, kernel_regularizer=l2(0.001))(GA_x)
GA_x = BatchNormalization()(GA_x)
GA_x = Activation('relu')(GA_x)
GA_x = Dense(1, 'sigmoid')(GA_x)
# ==================================================================================
Conv_creative_id = Conv1D(embedding_size, 15, 5,
activation='relu')(Embedded_fix_creative_id)
# ----------------------------------------------------------------------
Conv_GM_x = MaxPooling1D(7)(Conv_creative_id)
Conv_GM_x = Conv1D(embedding_size, 2, 1, activation='relu')(Conv_GM_x)
Conv_GM_x = GlobalMaxPooling1D()(Conv_GM_x)
Conv_GM_x = Dropout(0.5)(Conv_GM_x)
Conv_GM_x = Dense(embedding_size // 2,
kernel_regularizer=l2(0.001))(Conv_GM_x)
Conv_GM_x = BatchNormalization()(Conv_GM_x)
Conv_GM_x = Activation('relu')(Conv_GM_x)
Conv_GM_x = Dropout(0.5)(Conv_GM_x)
Conv_GM_x = Dense(embedding_size // 4,
kernel_regularizer=l2(0.001))(Conv_GM_x)
Conv_GM_x = BatchNormalization()(Conv_GM_x)
Conv_GM_x = Activation('relu')(Conv_GM_x)
Conv_GM_x = Dense(1, 'sigmoid')(Conv_GM_x)
# ----------------------------------------------------------------------
Conv_GA_x = AveragePooling1D(7)(Conv_creative_id)
Conv_GA_x = Conv1D(embedding_size, 2, 1, activation='relu')(Conv_GA_x)
Conv_GA_x = GlobalAveragePooling1D()(Conv_GA_x)
Conv_GA_x = Dropout(0.5)(Conv_GA_x)
Conv_GA_x = Dense(embedding_size // 2,
kernel_regularizer=l2(0.001))(Conv_GA_x)
Conv_GA_x = BatchNormalization()(Conv_GA_x)
Conv_GA_x = Activation('relu')(Conv_GA_x)
Conv_GA_x = Dropout(0.5)(Conv_GA_x)
Conv_GA_x = Dense(embedding_size // 4,
kernel_regularizer=l2(0.001))(Conv_GA_x)
Conv_GA_x = BatchNormalization()(Conv_GA_x)
Conv_GA_x = Activation('relu')(Conv_GA_x)
Conv_GA_x = Dense(1, 'sigmoid')(Conv_GA_x)
# ----------------------------------------------------------------------
LSTM_x = Conv1D(embedding_size, 14, 7, activation='relu')(Conv_creative_id)
LSTM_x = LSTM(embedding_size, return_sequences=True)(LSTM_x)
LSTM_x = LSTM(embedding_size, return_sequences=True)(LSTM_x)
LSTM_x = LSTM(embedding_size)(LSTM_x)
LSTM_x = Dropout(0.5)(LSTM_x)
LSTM_x = Dense(embedding_size // 2, kernel_regularizer=l2(0.001))(LSTM_x)
LSTM_x = BatchNormalization()(LSTM_x)
LSTM_x = Activation('relu')(LSTM_x)
LSTM_x = Dropout(0.5)(LSTM_x)
LSTM_x = Dense(embedding_size // 4, kernel_regularizer=l2(0.001))(LSTM_x)
LSTM_x = BatchNormalization()(LSTM_x)
LSTM_x = Activation('relu')(LSTM_x)
LSTM_x = Dense(1, 'sigmoid')(LSTM_x)
# ----------------------------------------------------------------------
concatenated = concatenate([
GM_x,
GA_x,
Conv_GM_x,
Conv_GA_x,
LSTM_x,
],
axis=-1)
output_tensor = Dense(1, 'sigmoid')(concatenated)
keras_api_model = Model(
[
# input_no_time_no_repeat,
Input_fix_creative_id,
# input_no_time_with_repeat,
],
output_tensor)
keras_api_model.summary()
plot_model(keras_api_model, to_file='model/keras_api_word2vec_model.png')
print('-' * 5 + ' ' * 3 + "编译模型" + ' ' * 3 + '-' * 5)
keras_api_model.compile(optimizer=optimizers.RMSprop(lr=RMSProp_lr),
loss=losses.binary_crossentropy,
metrics=[metrics.binary_accuracy])
return keras_api_model