def build_model(self):
premise = Sequential()
premise.add(
LSTM(input_shape=self.ds.X_p_train.shape[1:],
output_dim=ve.HIDDEN_SIZE))
premise.add(Dropout(ve.DROPOUT_PROB))
hypothesis = Sequential()
hypothesis.add(
LSTM(input_shape=self.ds.X_h_train.shape[1:],
output_dim=ve.HIDDEN_SIZE))
hypothesis.add(Dropout(ve.DROPOUT_PROB))
model = Sequential()
model.add(Merge([premise, hypothesis], mode=ve.CONCAT))
model.add(
Dense(output_dim=ve.HIDDEN_SIZE,
activation=ve.TANH,
W_regularizer=l2(self.hp.dense_l2)))
model.add(
Dense(output_dim=ve.NLI_CLASSES,
activation=ve.SOFTMAX,
W_regularizer=l2(self.hp.dense_l2)))
ve.compile_other_model(model, self.hp.optimizer)
return model
def build_model(self):
vector_input = Sequential()
vector_input.add(
Embedding(input_dim=len(self.ds.vocab),
output_dim=ve.EMBEDDING_SIZE,
weights=[self.ds.weights],
input_length=ve.WINDOW_SIZE))
caps_input = Sequential()
caps_input.add(
Embedding(input_dim=ve.CAPS_DIMS,
output_dim=ve.CAPS_DIMS,
weights=[np.eye(ve.CAPS_DIMS)],
input_length=ve.WINDOW_SIZE))
model = Sequential()
model.add(Merge([vector_input, caps_input], mode=ve.CONCAT))
model.add(
Reshape(((ve.EMBEDDING_SIZE + ve.CAPS_DIMS) * ve.WINDOW_SIZE, )))
model.add(Dense(output_dim=ve.HIDDEN_SIZE))
model.add(Activation(ve.TANH))
model.add(Dropout(ve.DROPOUT_PROB))
model.add(Dense(input_dim=ve.HIDDEN_SIZE, output_dim=ve.NER_CLASSES))
model.add(Activation(ve.SOFTMAX))
ve.compile_other_model(model, self.hp.optimizer)
return model
def build_model(self):
premise = Sequential()
premise.add(Embedding(input_dim=len(self.ds.vocab),
output_dim=ve.EMBEDDING_SIZE,
weights=[self.ds.weights],
input_length=ve.NLI_MAX_LEN,
W_regularizer=l2(self.hp.embedding_l2)))
premise.add(LSTM(output_dim=ve.HIDDEN_SIZE))
premise.add(Dropout(ve.DROPOUT_PROB))
hypothesis = Sequential()
hypothesis.add(Embedding(input_dim=len(self.ds.vocab),
output_dim=ve.EMBEDDING_SIZE,
weights=[self.ds.weights],
input_length=ve.NLI_MAX_LEN,
W_regularizer=l2(self.hp.embedding_l2)))
hypothesis.add(LSTM(input_shape=self.ds.X_h_train.shape[1:],
output_dim=ve.HIDDEN_SIZE))
hypothesis.add(Dropout(ve.DROPOUT_PROB))
model = Sequential()
model.add(Merge([premise, hypothesis],mode=ve.CONCAT))
model.add(Dense(output_dim=ve.HIDDEN_SIZE,activation=ve.TANH,
W_regularizer=l2(self.hp.dense_l2)))
model.add(Dense(output_dim=ve.NLI_CLASSES,activation=ve.SOFTMAX,
W_regularizer=l2(self.hp.dense_l2)))
ve.compile_other_model(model, self.hp.optimizer)
return model
def build_model(self):
model = Sequential()
model.add(Dense(input_shape=self.ds.X_train.shape[1:],
output_dim=ve.HIDDEN_SIZE,
W_regularizer=l2(self.hp.dense_l2)))
model.add(Activation(ve.TANH))
model.add(Dropout(ve.DROPOUT_PROB))
model.add(Dense(input_dim=ve.HIDDEN_SIZE,
output_dim=ve.POS_CLASSES,
W_regularizer=l2(self.hp.dense_l2)))
model.add(Activation(ve.SOFTMAX))
ve.compile_other_model(model, self.hp.optimizer)
return model
def build_model(self):
model = Sequential()
model.add(Embedding(input_dim=len(self.ds.vocab),
output_dim=ve.EMBEDDING_SIZE,
weights=[self.ds.weights],
input_length=ve.WINDOW_SIZE))
model.add(Reshape((ve.EMBEDDING_SIZE * ve.WINDOW_SIZE,)))
model.add(Dense(output_dim=ve.HIDDEN_SIZE))
model.add(Activation(ve.TANH))
model.add(Dropout(ve.DROPOUT_PROB))
model.add(Dense(input_dim=ve.HIDDEN_SIZE, output_dim=ve.CHUNK_CLASSES))
model.add(Activation(ve.SOFTMAX))
ve.compile_other_model(model, self.hp.optimizer)
return model
def build_model(self):
model = Sequential()
model.add(
Embedding(input_dim=len(self.ds.vocab),
output_dim=ve.EMBEDDING_SIZE,
weights=[self.ds.weights],
input_length=ve.QUESTIONS_MAX_LEN,
W_regularizer=l2(self.hp.embedding_l2)))
model.add(LSTM(output_dim=ve.HIDDEN_SIZE))
model.add(Dropout(ve.DROPOUT_PROB))
model.add(
Dense(input_dim=ve.HIDDEN_SIZE,
output_dim=ve.QUESTIONS_CLASSES,
W_regularizer=l2(self.hp.dense_l2)))
model.add(Activation(ve.SOFTMAX))
ve.compile_other_model(model, self.hp.optimizer)
return model