84 lines
2.4 KiB
Python
84 lines
2.4 KiB
Python
from keras.layers import Input, Dense, GRU, Dropout, BatchNormalization, MaxPooling1D, Conv1D, Flatten, Concatenate
|
|
from keras.models import Model
|
|
|
|
from coffeehouse_dltc.config import SAMPLE_LENGTH
|
|
|
|
|
|
def get_nn_model(nn_model, embedding, output_length):
|
|
if nn_model == 'cnn':
|
|
return cnn(embedding_size=embedding, output_length=output_length)
|
|
elif nn_model == 'rnn':
|
|
return rnn(embedding_size=embedding, output_length=output_length)
|
|
else:
|
|
raise ValueError("Unknown NN type: {}".format(nn_model))
|
|
|
|
|
|
# noinspection PyPep8Naming
|
|
def cnn(embedding_size, output_length):
|
|
""" Create and return a keras model of a CNN """
|
|
|
|
NB_FILTER = 256
|
|
NGRAM_LENGTHS = [1, 2, 3, 4, 5]
|
|
|
|
conv_layers, inputs = [], []
|
|
|
|
for ngram_length in NGRAM_LENGTHS:
|
|
current_input = Input(shape=(SAMPLE_LENGTH, embedding_size))
|
|
inputs.append(current_input)
|
|
|
|
convolution = Conv1D(
|
|
NB_FILTER,
|
|
ngram_length,
|
|
kernel_initializer='lecun_uniform',
|
|
activation='tanh',
|
|
)(current_input)
|
|
|
|
pool_size = SAMPLE_LENGTH - ngram_length + 1
|
|
pooling = MaxPooling1D(pool_size=pool_size)(convolution)
|
|
conv_layers.append(pooling)
|
|
|
|
merged = Concatenate()(conv_layers)
|
|
dropout = Dropout(0.5)(merged)
|
|
flattened = Flatten()(dropout)
|
|
outputs = Dense(output_length, activation='sigmoid')(flattened)
|
|
|
|
model = Model(inputs=inputs, outputs=outputs)
|
|
|
|
model.compile(
|
|
loss='binary_crossentropy',
|
|
optimizer='adam',
|
|
metrics=['top_k_categorical_accuracy'],
|
|
)
|
|
|
|
return model
|
|
|
|
|
|
def rnn(embedding_size, output_length):
|
|
""" Create and return a keras model of a RNN """
|
|
# noinspection PyPep8Naming
|
|
HIDDEN_LAYER_SIZE = 256
|
|
|
|
inputs = Input(shape=(SAMPLE_LENGTH, embedding_size))
|
|
|
|
gru = GRU(
|
|
HIDDEN_LAYER_SIZE,
|
|
input_shape=(SAMPLE_LENGTH, embedding_size),
|
|
kernel_initializer="glorot_uniform",
|
|
recurrent_initializer='normal',
|
|
activation='relu',
|
|
)(inputs)
|
|
|
|
batch_normalization = BatchNormalization()(gru)
|
|
dropout = Dropout(0.1)(batch_normalization)
|
|
outputs = Dense(output_length, activation='sigmoid')(dropout)
|
|
|
|
model = Model(inputs=inputs, outputs=outputs)
|
|
|
|
model.compile(
|
|
loss='binary_crossentropy',
|
|
optimizer='adam',
|
|
metrics=['top_k_categorical_accuracy'],
|
|
)
|
|
|
|
return model
|