-
Notifications
You must be signed in to change notification settings - Fork 46
/
config.py
131 lines (111 loc) · 5.76 KB
/
config.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
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
# -*- coding: utf-8 -*-
"""
@author: alexyang
@contact: alex.yang0326@gmail.com
@file: config.py
@time: 2019/1/5 9:59
@desc:
"""
class Config(object):
def __init__(self):
# input configuration
self.data_folder = 'laptop/term'
self.data_name = 'laptop'
self.level = 'word' # options are 'word' & 'char'
self.max_len = {'laptop': {'word': 83, 'char': 465}, 'restaurant': {'word': 79, 'char': 358},
'twitter': {'word': 73, 'char': 188}}
self.left_max_len = {'laptop': {'word': 70, 'char': 365}, 'restaurant': {'word': 72, 'char': 344},
'twitter': {'word': 39, 'char': 156}}
self.right_max_len = {'laptop': {'word': 78, 'char': 400}, 'restaurant': {'word': 72, 'char': 326},
'twitter': {'word': 67, 'char': 164}}
self.asp_max_len = {'laptop': {'word': 8, 'char': 58}, 'restaurant': {'word': 21, 'char': 115},
'twitter': {'word': 3, 'char': 21}}
self.word_embed_dim = 300
self.word_embed_trainable = False
self.word_embed_type = 'glove' # use what kind of word embeddings, can be pre-trained on a larger corpus or just on dataset
self.aspect_embed_dim = 300
self.aspect_embed_trainable = False
self.aspect_embed_type = 'glove' # use mean of word embeddings or just randomly initialization
self.use_text_input = None
self.use_text_input_l = None
self.use_text_input_r = None
self.use_text_input_r_with_pad = None
self.use_aspect_input = None
self.use_aspect_text_input = None
self.use_loc_input = None
self.use_offset_input = None
self.use_mask = None
self.is_aspect_term = True
# model structure configuration
self.exp_name = None
self.model_name = None
self.lstm_units = 300
self.dense_units = 128
# model training configuration
self.dropout = 0.2
self.batch_size = 32
self.n_epochs = 50
self.n_classes = 3
self.learning_rate = 0.001
self.optimizer = "adam"
# model saving configuration
self.checkpoint_dir = './ckpt'
self.checkpoint_monitor = 'val_acc'
self.checkpoint_save_best_only = True
self.checkpoint_save_weights_only = True
self.checkpoint_save_weights_mode = 'max'
self.checkpoint_verbose = 1
# early stopping configuration
self.early_stopping_monitor = 'val_acc'
self.early_stopping_patience = 5
self.early_stopping_verbose = 1
self.early_stopping_mode = 'max'
# elmo embedding configure
self.use_elmo = False
self.use_elmo_alone = False
self.elmo_hub_url = './raw_data/tfhub_elmo_2'
self.elmo_output_mode = 'elmo'
self.idx2token = None
self.idx2aspect_token = None
self.elmo_trainable = False
def init_input(self):
if self.model_name == 'td_lstm':
self.use_text_input, self.use_text_input_l, self.use_text_input_r = False, True, True
self.use_aspect_input, self.use_aspect_text_input = False, False
self.use_loc_input, self.use_offset_input = False, False
self.use_text_input_r_with_pad, self.use_mask = False, False
elif self.model_name == 'tc_lstm':
self.use_text_input, self.use_text_input_l, self.use_text_input_r = False, True, True
self.use_aspect_input, self.use_aspect_text_input = True, False
self.use_loc_input, self.use_offset_input = False, False
self.use_text_input_r_with_pad, self.use_mask = False, False
elif self.model_name in ['at_lstm', 'ae_lstm', 'atae_lstm']:
self.use_text_input, self.use_text_input_l, self.use_text_input_r = True, False, False
self.use_aspect_input, self.use_aspect_text_input = True, False
self.use_loc_input, self.use_offset_input = False, False
self.use_text_input_r_with_pad, self.use_mask = False, False
elif self.model_name == 'memnet':
self.use_text_input, self.use_text_input_l, self.use_text_input_r = True, False, False
self.use_aspect_input, self.use_aspect_text_input = True, False
self.use_loc_input, self.use_offset_input = True, False
self.use_text_input_r_with_pad, self.use_mask = False, False
elif self.model_name == 'ram':
self.use_text_input, self.use_text_input_l, self.use_text_input_r = True, False, False
self.use_aspect_input, self.use_aspect_text_input = True, False
self.use_loc_input, self.use_offset_input = True, True
self.use_text_input_r_with_pad, self.use_mask = False, False
elif self.model_name == 'ian':
self.use_text_input, self.use_text_input_l, self.use_text_input_r = True, False, False
self.use_aspect_input, self.use_aspect_text_input = False, True
self.use_loc_input, self.use_offset_input = False, False
self.use_text_input_r_with_pad, self.use_mask = False, False
elif self.model_name == 'cabasc':
self.use_text_input, self.use_text_input_l, self.use_text_input_r = True, True, False
self.use_aspect_input, self.use_aspect_text_input = True, False
self.use_loc_input, self.use_offset_input = False, False
self.use_text_input_r_with_pad, self.use_mask = True, True
else:
raise ValueError('model name `{}` not understood'.format(self.model_name))
if not self.is_aspect_term:
self.use_loc_input = False
self.use_offset_input = False