max_len = 400
embedding_size = 32
vocab_size = 20000
head_num = 8
hopping_num = 1
max_epoch = 20
l2_penalty_coef = 1e-4
x_train, x_test, y_train, y_test = load_imdb(vocab_size)
for i, sentence in enumerate(tqdm(x_train)):
x_train[i] = [vocab_size] + sentence
for i, sentence in enumerate(tqdm(x_test)):
x_test[i] = [vocab_size] + sentence
x_train = with_padding(x_train,
padding_type='post',
max_sequence_length=max_len)
x_test = with_padding(x_test, padding_type='post', max_sequence_length=max_len)
y_train = y_train[:, None]
y_test = y_test[:, None]
num_train_batch = len(x_train) // batch_size
num_dev_batch = len(x_test) // batch_size
def load_train_func(index):
return x_train[index], y_train[index]
def load_dev_func(index):
return x_test[index], y_test[index]
from functions import time_distributed
from functions import time_distributed_softmax_cross_entropy
"""cuda setting"""
from nnabla.contrib.context import extension_context
ctx = extension_context('cuda.cudnn', device_id=0)
nn.set_default_context(ctx)
""""""
# nn.load_parameters('encdec_best.h5')
from utils import load_data
from utils import with_padding
train_source, dev_source, test_source, w2i_source, i2w_source = load_data(
'./data', 'en')
train_source = with_padding(train_source,
padding_type='post')[:, ::-1].astype(np.int32)
dev_source = with_padding(dev_source,
padding_type='post')[:, ::-1].astype(np.int32)
test_source = with_padding(test_source,
padding_type='post')[:, ::-1].astype(np.int32)
train_target, dev_target, test_target, w2i_target, i2w_target = load_data(
'./data', 'ja')
train_target = with_padding(train_target, padding_type='post').astype(np.int32)
dev_target = with_padding(dev_target, padding_type='post').astype(np.int32)
test_target = with_padding(test_target, padding_type='post').astype(np.int32)
vocab_size_source = len(w2i_source)
vocab_size_target = len(w2i_target)
sentence_length_source = train_source.shape[1]
sentence_length_target = train_target.shape[1]
from parametric_functions import simple_rnn
from functions import time_distributed
from functions import time_distributed_softmax_cross_entropy
"""cuda setting"""
from nnabla.contrib.context import extension_context
ctx = extension_context('cuda.cudnn', device_id=1)
nn.set_default_context(ctx)
""""""
from utils import load_data
from utils import w2i, i2w, c2i, i2c, word_length
from utils import with_padding
train_data = load_data('./ptb/train.txt')
train_data = with_padding(train_data, padding_type='post')
valid_data = load_data('./ptb/valid.txt')
valid_data = with_padding(valid_data, padding_type='post')
vocab_size = len(w2i)
sentence_length = 20
embedding_size = 128
hidden_size = 128
batch_size = 256
max_epoch = 100
x_train = train_data[:, :sentence_length].astype(np.int32)
y_train = train_data[:, 1:sentence_length + 1].astype(np.int32)
x_valid = valid_data[:, :sentence_length].astype(np.int32)