from rnn import *
import data
use_gpu(0) # -1:cpu; 0,1,2,..: gpu
e = 0.0
lr = 0.5
drop_rate = 0.
batch_size = 1000
hidden_size = [500]
# try: gru, lstm
cell = "gru"
# try: sgd, momentum, rmsprop, adagrad, adadelta, adam
optimizer = "adadelta"
seqs, i2w, w2i, data_xy = data.word_sequence("/data/toy.txt", batch_size)
dim_x = len(w2i)
dim_y = len(w2i)
num_sents = data_xy[0][3]
print "#features = ", dim_x, "#labels = ", dim_y
print "compiling..."
model = RNN(dim_x, dim_y, hidden_size, cell, optimizer, drop_rate, num_sents)
print "training..."
start = time.time()
g_error = 9999.9999
for i in xrange(2000):
error = 0.0
in_start = time.time()
for batch_id, xy in data_xy.items():
from rnn import *
import data
use_gpu(0) # -1:cpu; 0,1,2,..: gpu
e = 0.0
lr = 0.5
drop_rate = 0.
batch_size = 1000
hidden_size = [500]
# try: gru, lstm
cell = "gru"
# try: sgd, momentum, rmsprop, adagrad, adadelta, adam
optimizer = "adadelta"
seqs, i2w, w2i, data_xy = data.word_sequence("/data/toy.txt", batch_size)
dim_x = len(w2i)
dim_y = len(w2i)
num_sents = data_xy[0][3]
print "#features = ", dim_x, "#labels = ", dim_y
print "compiling..."
model = RNN(dim_x, dim_y, hidden_size, cell, optimizer, drop_rate, num_sents)
print "training..."
start = time.time()
g_error = 9999.9999
for i in xrange(2000):
error = 0.0
in_start = time.time()
for batch_id, xy in data_xy.items():
print "initializing model"
model = RNN(embedding_size, embedding_size, hidden_size)
for _ in xrange(epoch):
print "new epoch"
error = 0.0
in_start = time.time()
for i in xrange(number_of_batch):
print "batch: ", i
training_files = files[i*batch_size : (i+1)*batch_size]
for file in training_files:
print "loading data from file: ", file
data_x, data_y, w2v_model, unknown_word_vector = data.word_sequence(os.path.join(training_folder, file),
w2v_model,
unknown_word_vector,
embedding_size)
X, mask_X = data_x
Y, mask_Y = data_y
print "training"
cost, sents = model.train(X, mask_X, Y, mask_Y, lr)
error += cost
print "decoding to text"
for s in xrange(int(sents.shape[1] / embedding_size)):
xs = sents[:, s * embedding_size : (s + 1) * embedding_size]
for w_i in xrange(xs.shape[0]):
vector = xs[w_i, :]
w, score = w2v_model.similar_by_vector(vector, topn=1)[0]
import data
import test_data
def predict():
pass
e = 0.01 #error
lr = 0.5
drop_rate = 0.
batch_size = 1 #must be same as the loading model's batch size
hidden_size = [500]
# try: gru, lstm
cell = "gru"
# try: sgd, momentum, rmsprop, adagrad, adadelta, adam
optimizer = "adadelta"
test_path = "data/test.txt"
test_data = data.word_sequence(test_path, batch_size, left)
i2w, w2i = load_data_dic("data/i2w.pkl", "data/w2i.pkl")
dim_x = len(w2i)
dim_y = len(w2i)
num_sents = batch_size
parser.add_argument("--model")
args = parser.parse_args(sys.argv[1:])
use_gpu(args.gpu) # -1:cpu; 0,1,2,..: gpu
e = 0.01
lr = 0.1
drop_rate = 0.
batch_size = args.bsize
hidden_size = [args.hid]
# try: gru, lstm
cell = "lstm"
# try: sgd, momentum, rmsprop, adagrad, adadelta, adam, nesterov_momentum
optimizer = "adadelta"
seqs, i2w, w2i, data_xy = data.word_sequence(args.path, batch_size)
dim_x = len(w2i)
dim_y = len(w2i)
num_sents = data_xy[0][3]
print "#features = ", dim_x, "#labels = ", dim_y
print "compiling..."
model = RNN(dim_x, dim_y, hidden_size, cell, optimizer, drop_rate, num_sents)
print "training..."
start = time.time()
g_error = 9999.9999
if args.model is None:
for i in xrange(args.epoch):
error = 0.0
in_start = time.time()
for batch_id, xy in data_xy.items():
print "initializing model"
model = RNN(embedding_size, embedding_size, hidden_size)
for _ in xrange(epoch):
print "new epoch"
error = 0.0
in_start = time.time()
for i in xrange(number_of_batch):
print "batch: ", i
training_files = files[i * batch_size:(i + 1) * batch_size]
for file in training_files:
print "loading data from file: ", file
data_x, data_y, w2v_model, unknown_word_vector = data.word_sequence(
os.path.join(training_folder, file), w2v_model,
unknown_word_vector, embedding_size)
X, mask_X = data_x
Y, mask_Y = data_y
print "training"
cost, sents = model.train(X, mask_X, Y, mask_Y, lr)
error += cost
print "decoding to text"
for s in xrange(int(sents.shape[1] / embedding_size)):
xs = sents[:, s * embedding_size:(s + 1) * embedding_size]
for w_i in xrange(xs.shape[0]):
vector = xs[w_i, :]
w, score = w2v_model.similar_by_vector(vector, topn=1)[0]
