|
1
|
|
|
import os |
|
2
|
|
|
import logging |
|
3
|
|
|
from argparse import ArgumentParser |
|
4
|
|
|
|
|
5
|
|
|
from vocab import Vocab |
|
6
|
|
|
from lmdataset import LMDataset |
|
7
|
|
|
from lm import NeuralLM |
|
8
|
|
|
from deepy.dataset import SequentialMiniBatches |
|
9
|
|
|
from deepy.trainers import SGDTrainer, LearningRateAnnealer |
|
10
|
|
|
from deepy.layers import IRNN, Dense |
|
11
|
|
|
|
|
12
|
|
|
|
|
13
|
|
|
logging.basicConfig(level=logging.INFO) |
|
14
|
|
|
|
|
15
|
|
|
resource_dir = os.path.abspath(os.path.dirname(__file__)) + os.sep + "resources" |
|
16
|
|
|
|
|
17
|
|
|
vocab_path = os.path.join(resource_dir, "ptb.train.txt") |
|
18
|
|
|
train_path = os.path.join(resource_dir, "ptb.train.txt") |
|
19
|
|
|
valid_path = os.path.join(resource_dir, "ptb.valid.txt") |
|
20
|
|
|
vocab = Vocab(char_based=True) |
|
21
|
|
|
vocab.load(vocab_path, max_size=1000) |
|
22
|
|
|
|
|
23
|
|
|
model = NeuralLM(input_dim=vocab.size, input_tensor=3) |
|
24
|
|
|
model.stack( |
|
25
|
|
|
IRNN(hidden_size=100, output_type="sequence"), |
|
26
|
|
|
Dense(vocab.size, "softmax")) |
|
27
|
|
|
|
|
28
|
|
|
|
|
29
|
|
|
if __name__ == '__main__': |
|
|
|
|
|
|
30
|
|
|
ap = ArgumentParser() |
|
31
|
|
|
ap.add_argument("--model", default=os.path.join(os.path.dirname(__file__), "models", "char_irnn_model1.gz")) |
|
32
|
|
|
ap.add_argument("--sample", default="") |
|
33
|
|
|
args = ap.parse_args() |
|
34
|
|
|
|
|
35
|
|
|
if os.path.exists(args.model): |
|
36
|
|
|
model.load_params(args.model) |
|
37
|
|
|
|
|
38
|
|
|
lmdata = LMDataset(vocab, train_path, valid_path, history_len=30, char_based=True, max_tokens=300) |
|
39
|
|
|
batch = SequentialMiniBatches(lmdata, batch_size=20) |
|
40
|
|
|
|
|
41
|
|
|
trainer = SGDTrainer(model) |
|
42
|
|
|
annealer = LearningRateAnnealer(trainer) |
|
43
|
|
|
|
|
44
|
|
|
trainer.run(batch, controllers=[annealer]) |
|
45
|
|
|
|
|
46
|
|
|
model.save_params(args.model) |
|
47
|
|
|
|
uaca /
deepy
Shu
Duplicated code is one of the most pungent code smells. If you need to duplicate the same code in three or more different places, we strongly encourage you to look into extracting the code into a single class or operation.
You can also find more detailed suggestions in the “Code” section of your repository.