-
Notifications
You must be signed in to change notification settings - Fork 6
/
qe_hyperparams.py
57 lines (47 loc) · 2.17 KB
/
qe_hyperparams.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
# -*- coding: utf-8 -*-
#/usr/bin/python3
'''
June 2017 by kyubyong park.
kbpark.linguist@gmail.com.
https://www.github.com/kyubyong/transformer
maxlen = 70, batch_size = 128
源端、目标端、proj 词表共享 ???qe-brain中没有共享
优化策略:已经更改为和qe-brain中相同
'''
class QE_Hyperparams:
'''Hyperparameters'''
# data
source_train = '../parallel_data/2017/qe_data/sentence_level_en_de/train.tok.lower.src'
target_train = '../parallel_data/2017/qe_data/sentence_level_en_de/train.tok.lower.mt'
label_train = '../parallel_data/2017/qe_data/sentence_level_en_de/train.hter'
source_dev = '../parallel_data/2017/qe_data/sentence_level_en_de/dev.tok.lower.src'
target_dev = '../parallel_data/2017/qe_data/sentence_level_en_de/dev.tok.lower.mt'
label_dev = '../parallel_data/2017/qe_data/sentence_level_en_de/dev.hter'
source_test = '../parallel_data/2017/qe_data/sentence_level_en_de_test/test.2017.tok.lower.src'
target_test = '../parallel_data/2017/qe_data/sentence_level_en_de_test/test.2017.tok.lower.mt'
label_test = '../parallel_data/2017/qe_data/sentence_level_en_de_test/en-de_task1_test.2017.hter'
vocab_dir = './preprocessed_qe/'
pattern = 'en-de'
# training
batch_size = 64 # alias = N
lr = 2.0 # learning rate. learning rate is adjusted to the global step.
warmup_steps = 8000
log_dir = 'logdir' # log directory,save expert_model
num_keep_ckpts = 5
# model
maxlen = 70+2 # Maximum number of words in a sentence. alias = T.
# Feel free to increase this if you are ambitious.
#min_cnt = 20 # words whose occurred less than min_cnt are encoded as <UNK>.
vocab_size = 30000 # src and tgt
hidden_units = 512 # alias = C
num_blocks = 2 # number of encoder/decoder blocks
num_heads = 8
dropout_rate = 0.1
sinusoid = True # If True, use sinusoid. If false, positional embedding.
# qe params
model_dir = './modeldir/' # dir of qe_model
num_train_steps = 75000
steps_per_stats = 10 # Once in a while, we print statistics.
steps_per_save = 50
fixed_exp = False # fixed expert weights or not
patience = 5