def main():
data_root = '../data/wordinds/'
print 'starting L matrix construction'
gw = GloveWrapper(verbose=True)
L0 = gw.L
train_files = os.listdir(data_root+ '/train/')
print 'getting train and test data'
train_x, train_y, train_D, _ = get_data(gw, train_files, 'train/')
#pdb.set_trace()
test_x, test_y, test_D, _ = get_data(gw, train_files, 'test/')
D0 = np.random.randn(train_D[-1] + 1, 300)
print 'got train and test data'
n_epochs = 25
train_x = train_x*n_epochs
train_y = train_y*n_epochs
train_D = train_D*n_epochs
model = DRNNLM(L0, D0, U0 = L0, alpha=0.05, rseed=10, bptt=3)
print 'constructed model, training...'
model.custom_train_sgd(train_x,train_y, train_D, apply_to=['H','U','L','D'], printevery=5, costevery=25)
print 'training done'
print 'saving model'
with open('../data/drnnlm_model.pkl', 'w') as model_file:
pkl.dump(model, model_file)
print 'model saved'
for i in range(10):
seq, J = model.generate_sequence(1, gw.get_index("SSTART"), gw.get_index("EEND"), maxlen=100)
print " ".join(gw, seq_to_words(seq))
def main():
data_root = '../data/wordinds/'
print 'starting L matrix construction'
gw = GloveWrapper(verbose=True)
L0 = gw.L
train_files = os.listdir(data_root+ '/train/')
print 'getting train and test data'
train_x, train_y, train_D, _ = get_data(gw, train_files, 'train/')
pdb.set_trace()
test_x, test_y, test_D, _ = get_data(gw, train_files, 'test/')
D0 = np.random.randn(train_D[-1] + 1, 300)
print 'got train and test data'
n_epochs = 25
train_x = train_x*n_epochs
train_y = train_y*n_epochs
train_D = train_D*n_epochs
model = SimpleDRNNLM(L0, D0, U0 = L0, alpha=0.05, rseed=10, bptt=3)
print 'constructed model, training...'
model.custom_train_sgd(train_x,train_y, train_D, apply_to=['H','U','L','D'], printevery=5, costevery=25)
print 'training done'
print 'saving model'
with open('../data/simple_drnnlm_model.pkl', 'w') as model_file:
pkl.dump(model, model_file)
print 'model saved'
for i in range(10):
seq, J = model.generate_sequence(1, gw.get_index("SSTART"), gw.get_index("EEND"), maxlen=100)
print " ".join(seq_to_words(seq))
def main():
data_root = "../data/wordinds/"
print "starting L matrix construction"
gw = GloveWrapper(verbose=True)
L0 = gw.L
train_files = os.listdir(data_root + "/train/")
print "getting train and test data"
train_x, train_y, train_D, _ = get_data(gw, train_files, "train/")
pdb.set_trace()
test_x, test_y, test_D, _ = get_data(gw, train_files, "test/")
D0 = np.random.randn(train_D[-1] + 1, 300)
print "got train and test data"
n_epochs = 25
train_x = train_x * n_epochs
train_y = train_y * n_epochs
train_D = train_D * n_epochs
model = SimpleDRNNLM(L0, D0, U0=L0, alpha=0.05, rseed=10, bptt=3)
print "constructed model, training..."
model.custom_train_sgd(train_x, train_y, train_D, apply_to=["H", "U", "L", "D"], printevery=5, costevery=25)
print "training done"
print "saving model"
with open("../data/simple_drnnlm_model.pkl", "w") as model_file:
pkl.dump(model, model_file)
print "model saved"
for i in range(10):
seq, J = model.generate_sequence(1, gw.get_index("SSTART"), gw.get_index("EEND"), maxlen=100)
print " ".join(seq_to_words(seq))
class TextPreprocessor():
def __init__(self):
self.glove_vecs = GloveWrapper(verbose=True)
def doc_to_inds(self, doc):
doc = doc.lower()
doc = doc.replace('\n', ' ')
inds = []
for sent in nltk.tokenize.sent_tokenize(doc):
words = nltk.word_tokenize(sent)
sent_of_inds = []
sent_of_inds.append(self.glove_vecs.get_index('SSTART'))
for word in words:
sent_of_inds.append(self.glove_vecs.get_index(word))
sent_of_inds.append(self.glove_vecs.get_index('EEND'))
inds.append(sent_of_inds)
return inds