def get_keras_with_word2vec(name, data_dto, extra_param=None, verbose=False):
vectorize_data = Vectorized(data_dto)
vectorize_data.initialize_with_word2vec()
model, path = get_saved_model(name)
if (model == None):
model = Sequential()
model.add(Dense(32, activation='relu', input_dim=data_dto.vocab_size))
model.add(
Dense(len(vectorize_data.data_dto.target_names),
activation='softmax'))
model.compile(optimizer='rmsprop',
loss='binary_crossentropy',
metrics=['accuracy'])
if (verbose):
model.summary()
model.fit(vectorize_data.X_train,
vectorize_data.y_train,
epochs=9,
batch_size=32,
validation_data=(vectorize_data.X_test,
vectorize_data.y_test),
verbose=verbose)
model.save(path)
return KerasClassifierWithWord2Vec(name, model, vectorize_data, verbose)
def get_simple_keras_classifier(name, data_dto, count_vectorizer_dto=None, verbose=False):
if count_vectorizer_dto == None:
count_vectorizer_dto = CountVectorizerDTO()
vectorize_data = Vectorized(data_dto)
vectorize_data.initialize_with_count_vectorizer(count_vectorizer_dto)
model, path = get_saved_model(name)
if (model == None):
model = Sequential()
model.add(Dense(10, input_dim=vectorize_data.input_dim, activation='relu'))
model.add(Dense(len(vectorize_data.data_dto.target_names), activation='softmax', kernel_initializer='uniform'))
# categorial is the way to go for multiple possible categorys as results, instead of binary
model.compile(loss='categorical_crossentropy',
optimizer='adam',
metrics=['accuracy'])
if (verbose):
model.summary()
history = model.fit(vectorize_data.X_train, vectorize_data.y_train,
epochs=100,
verbose=verbose,
validation_data=(vectorize_data.X_test, vectorize_data.y_test),
batch_size=100)
model.save(path)
if (verbose):
print (history)
return KerasClassifierWithVectorizer(name, model, vectorize_data, verbose)
def get_denser_keras_classifier_with_tokenizer(name,
data_dto,
keras_tokenizer_dto=None,
verbose=False):
if (keras_tokenizer_dto == None):
keras_tokenizer_dto = KerasTokenizerDTO()
vectorize_data = Vectorized(data_dto)
vectorize_data.initialize_with_keras_tokenizer(keras_tokenizer_dto)
model, path = get_saved_model(name)
if (model == None):
model = Sequential()
model.add(
Dense(512, input_shape=(vectorize_data.data_dto.vocab_size, )))
model.add(Activation('relu'))
model.add(Dropout(0.3))
model.add(Dense(512))
model.add(Activation('relu'))
model.add(Dropout(0.3))
model.add(Dense(len(vectorize_data.data_dto.target_names)))
model.add(Activation('softmax'))
# categorial is the way to go for multiple possible categorys as results, instead of binary
model.compile(loss='categorical_crossentropy',
optimizer='adam',
metrics=['accuracy'])
if (verbose):
model.summary()
history = model.fit(vectorize_data.X_train,
vectorize_data.y_train,
epochs=30,
batch_size=100,
verbose=verbose,
validation_split=0.1)
model.save(path)
if (verbose):
print(history)
return KerasClassifierWithTokenizer(name, model, vectorize_data, verbose)