def trainWord2Vec(latentFactorN = 10, weightSave = './TrainingTemp/Word2vecWeight.npy', epoch=20):
'''
Train a word2vec model based on Shakespear's sonnet, and save it's weight in TrainingTemp folder
'''
encodedSonnets, encodedSyllaDict, code2word, punc2code = encodedShake()
X, y = gnerateEmbedTrain(encodedSonnets)
model = Sequential()
model.add(Dense(units=latentFactorN, input_dim=len(code2word)))
model.add(Dense(units=len(code2word), activation='softmax'))
model.compile(loss='categorical_crossentropy', optimizer='adam', metrics=['accuracy'])
model.summary()
model.fit(to_categorical(X), to_categorical(y), epochs=epoch, verbose=1)
weights = model.layers[0].get_weights()[0]
weights = weights / np.max(np.abs(weights))
np.save(weightSave, weights)
return weights
import numpy as np
from HMM_fast import HiddenMarkovModel, unsupervised_HMM
from DataProcessing import encodedShake, Convert2SonnetNaive
if __name__ == '__main__':
encodedSonnets, encodedSyllaDict, code2word, punc2code = encodedShake()
HMmodel = unsupervised_HMM(encodedSonnets, 40, 200)
print(Convert2SonnetNaive(encodedSonnets[0], code2word)[0])
print(Convert2SonnetNaive(encodedSonnets[0], code2word)[1])
sonnet = HMmodel.generate_emission(300)[0]
print(Convert2SonnetNaive(sonnet, code2word)[1])
np.save('./TrainingTemp/HMM.A.npy', HMmodel.A)
np.save('./TrainingTemp/HMM.O.npy', HMmodel.O)