def model1(maxlen, batch_size, num_epochs, w2v, traindf, cvdf):
train_gen = batch_generator(
df=traindf,
encoder=lambda b: data_init.encode_w2v(df=b, w2v=w2v, maxlen=maxlen),
batch_size=batch_size,
force_batch_size=True)
cv_gen = batch_generator(
df=cvdf,
encoder=lambda b: data_init.encode_w2v(df=b, w2v=w2v, maxlen=maxlen),
batch_size=batch_size)
# creates the neural network
model = Sequential()
model.add(LSTM(60, input_dim=300, return_sequences=True))
model.add(Dropout(0.5))
model.add(LSTM(60))
model.add(Dropout(0.5))
model.add(Dense(1, activation='sigmoid'))
# compiles the model
model.compile('rmsprop',
'binary_crossentropy',
metrics=['accuracy', 'mse'])
nb_val_samples = len(cvdf)
return model, train_gen, cv_gen, nb_val_samples
def model2(maxlen, batch_size, num_epochs, w2v, traindf, cvdf):
def encoder(b):
encoded_x, encoded_y = data_init.encode_w2v(df=b,
w2v=w2v,
maxlen=maxlen)
return [[encoded_x, encoded_x, encoded_x], encoded_y]
train_gen = batch_generator(df=traindf,
encoder=encoder,
batch_size=batch_size,
force_batch_size=True)
cv_gen = batch_generator(df=cvdf, encoder=encoder, batch_size=batch_size)
# creates the neural network consisting in 3 convolutional layers going to an LSTM layer
model1 = Sequential()
model1.add(
Convolution1D(50,
6,
border_mode='valid',
activation='relu',
input_shape=(maxlen, 300)))
model1.add(MaxPooling1D(pool_length=3, border_mode='valid'))
model1.add(Dropout(0.5))
model2 = Sequential()
model2.add(
Convolution1D(50,
5,
border_mode='valid',
activation='relu',
input_shape=(maxlen, 300)))
model2.add(MaxPooling1D(pool_length=3, border_mode='valid'))
model2.add(Dropout(0.5))
model3 = Sequential()
model3.add(
Convolution1D(50,
4,
border_mode='valid',
activation='relu',
input_shape=(maxlen, 300)))
model3.add(MaxPooling1D(pool_length=3, border_mode='valid'))
model3.add(Dropout(0.5))
model = Sequential()
model.add(Merge([model1, model2, model3], mode='concat', concat_axis=1))
model.add(LSTM(60, return_sequences=True))
model.add(Dropout(0.5))
model.add(LSTM(60))
model.add(Dropout(0.5))
model.add(Dense(1, activation='sigmoid'))
# compiles the model
model.compile('adam', 'binary_crossentropy', metrics=['accuracy', 'mse'])
nb_val_samples = len(cvdf)
return model, train_gen, cv_gen, nb_val_samples
def get_model_preds(filename, testdf, maxlen, batch_size, encoder):
model = load_model(filename)
print 'Getting regression predictions for %s...' % filename
test_gen = data_init.batch_generator(df=testdf,
encoder=encoder,
batch_size=batch_size,
shuffle=False)
return model.predict_generator(test_gen, val_samples=len(testdf))
def model5(maxlen,
batch_size,
num_epochs,
alphabet,
traindf,
cvdf,
categorical=False):
def encoder(b):
encoded_x, encoded_y = data_init.encode_embed(df=b,
alphabet=alphabet,
maxlen=maxlen,
categorical=categorical)
return [[encoded_x, encoded_x, encoded_x], encoded_y]
train_gen = balanced_batch_generator(df=traindf,
encoder=encoder,
batch_size=batch_size,
force_batch_size=True)
cv_gen = batch_generator(df=cvdf,
encoder=encoder,
batch_size=batch_size,
shuffle=False)
# creates the neural network consisting in 3 convolutional layers going into an LSTM layer
inputs = []
for filter_size in (6, 5, 4):
model1 = Sequential()
model1.add(Embedding(30000, 100, input_length=maxlen))
model1.add(
Convolution1D(310,
filter_size,
border_mode='valid',
activation='relu'))
model1.add(MaxPooling1D(pool_length=3, border_mode='valid'))
model1.add(Dropout(0.5))
model1.add(
Convolution1D(100, 3, border_mode='valid', activation='relu'))
model1.add(MaxPooling1D(pool_length=3, border_mode='valid'))
model1.add(Dropout(0.5))
model1.add(LSTM(60, return_sequences=True))
model1.add(Dropout(0.5))
inputs.append(model1)
model = Sequential()
model.add(Merge(inputs, mode='sum', concat_axis=1))
model.add(LSTM(60))
model.add(Dropout(0.5))
if categorical:
model.add(Dense(5, activation='softmax'))
model.compile('rmsprop',
'categorical_crossentropy',
metrics=['accuracy', 'mse'])
else:
model.add(Dense(1, activation='sigmoid'))
model.compile('adam',
'binary_crossentropy',
metrics=['accuracy', 'mse'])
nb_val_samples = len(cvdf)
return model, train_gen, cv_gen, nb_val_samples