sacar/cli.py at master · r-mal/sacar · GitHub

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
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
import argparse
import sys
import json
import os
from collections import defaultdict

import train
from data import dataset, pretrain
import visualization

run_switcher = {
    'train': train.main,
    'preprocess': dataset.preprocess,
    'eval': None,
    'viz': visualization.save_act_data,
    'pretrain-create': pretrain.preprocess,
    'pretrain': train.pretrain
  }


class AttrInfo(object):
  def __init__(self, attr_dir):
    print('Loading attribute info from %s...' % attr_dir)
    self.attr_label_maps = {}
    self.attr_weight_maps = {}
    self.common_attributes = {'polarity', 'modality'}
    self.activity_attributes = []
    for fname in os.listdir(attr_dir):
      if fname.endswith('.json'):
        attr = fname.replace('.json', '')
        self.attr_label_maps[attr] = json.load(open(os.path.join(attr_dir, fname)))
        if attr not in self.common_attributes:
          self.activity_attributes += [attr]
    weight_dir = os.path.join(attr_dir, 'attr_class_weights')
    for fname in os.listdir(weight_dir):
      if fname.endswith('.json'):
        attr = fname.replace('.json', '')
        self.attr_weight_maps[attr] = defaultdict(lambda: 0.)
        for k, v in json.load(open(os.path.join(weight_dir, fname))).items():
          self.attr_weight_maps[attr][k] = v
    self.activity_attributes = set(self.activity_attributes)


class DataInfo(object):
  def __init__(self, data_dir):
    self.record_dir = data_dir
    self.json_data = os.path.join(data_dir, 'data.json')
    self.word2id = json.load(open(os.path.join(data_dir, 'word2id.json')))
    self.boundary2id = json.load(open(os.path.join(data_dir, 'boundary2id.json')))
    self.rel2id = json.load(open(os.path.join(data_dir, 'rel2id.json')))


def load_or_default(field, default, loader=lambda x: x):
  # returns the loaded field if it is set or the default otherwise
  if field is None:
    return default
  else:
    return loader(field)


# noinspection PyTypeChecker
def main():
  parser = argparse.ArgumentParser(
    description='Train and Evaluate the Joint learning network for EEG knowledge extraction')

  # control options
  parser.add_argument('mode', choices=run_switcher.keys())
  parser.add_argument('--run_name', default=None, help='Name for the run')
  parser.add_argument('--verbose_eval', default=False, type=bool,
                      help='Print fine-grained evaluations (per class in addition to per-task)')
  parser.add_argument('--hard_eval', default=False, type=bool,
                      help='Evaluate attr/relation with predicted concept spans')
  parser.add_argument('--validation_proportion', default=0.2, type=float)
  parser.add_argument('--warm_start_model', default=None, help='model directory with saved model to use for warm start')
  parser.add_argument('--embedding_device', default='/device:GPU:0',
                      choices=['/device:GPU:0', '/device:GPU:1', '/cpu:0'])
  parser.add_argument('--pretrained', default=False, type=bool)

  # data
  parser.add_argument('--model_dir', default='/home/rmm120030/working/eeg/joint/model', help='model dir')
  parser.add_argument('--base_data_dir', default='/home/rmm120030/working/eeg/joint/data/io/',
                      help='Base data directory', type=DataInfo)
  parser.add_argument('--record_dir', default=None, help=('Location of directory the tfrecords should be in.' +
                                                          'If None, will use base_data_dir/data.json'))
  parser.add_argument('--train_filename', default='train.tfrecord', help='Train tfrecord filename')
  parser.add_argument('--val_filename', default='val.tfrecord', help='Validation tfrecord filename')
  parser.add_argument('--json_data', default=None, help=('Location of raw json data.' +
                                                         'If None, will use base_data_dir/data.json'))
  parser.add_argument('--word2id', default=None, help=('Json file with word to id mapping.' +
                                                       'If None, will use base_data_dir/word2id.json'))
  parser.add_argument('--boundary2id', default=None, help=('Json file with boundary_label to id mapping. ' +
                                                           'If None, will use base_data_dir/boundary2id.json'))
  parser.add_argument('--rel2id', default=None, help=('Json file with relation_label to id mapping. ' +
                                                      'If None, will use base_data_dir/rel2id.json'))
  parser.add_argument('--attr_info', default='/home/rmm120030/working/eeg/joint/data/attr', type=AttrInfo,
                      help='Directory with attribute to id mapping files.')
  parser.add_argument('--pretrain_data_dir', default='/home/rmm120030/working/eeg/joint/data/pretrain')

  # hyperparameters
  parser.add_argument('--learning_rate', default=1e-4, type=float)
  parser.add_argument('--batch_size', default=4, type=int)
  parser.add_argument('--drop_prob', default=0.1, type=float)
  parser.add_argument('--boundary_loss_param', default=1, type=float)
  parser.add_argument('--act_loss_param', default=0.1, type=float)
  parser.add_argument('--attr_loss_param', default=1, type=float)
  parser.add_argument('--rel_loss_param', default=1, type=float)
  parser.add_argument('--num_epochs', default=100, type=int)
  parser.add_argument('--steps_per_epoch', default=10, type=int)
  parser.add_argument('--class_weighting', default=False, type=bool)
  parser.add_argument('--multi_layer_attr', default=False, type=bool)
  parser.add_argument('--eval_delay', default=180, type=int)
  parser.add_argument('--eval_throttle', default=600, type=int)

  # parameters
  parser.add_argument('--modules', default=['concept', 'attr', 'relation'], nargs='+',
                      help='Modules to include: [concept, attr, relation]')
  parser.add_argument('--max_doc_len', default=519, type=int, help='Maximum number of tokens in a document')
  parser.add_argument('--vocab_size', default=21078, type=int, help='Number of unique tokens')  # 21074
  parser.add_argument('--embedding_size', default=256, type=int, help='Word embedding size')
  parser.add_argument('--base_transformer_num_heads', default=8, type=int,
                      help='Number of attention heads in the base transformer')
  parser.add_argument('--base_transformer_num_stacks', default=6, type=int,
                      help='Number of times the base transformer is stacked')
  parser.add_argument('--use_convolution', default=True, type=bool, help='Use convolutional layer in transformers?')
  parser.add_argument('--transformer_type', default='adaptive', choices=['transformer', 'universal', 'adaptive'],
                      help='transformer type for narrative/sentence encoders')
  # attr
  parser.add_argument('--sents_per_doc', default=25, type=int, help='Maximum number of sentences in a document')
  parser.add_argument('--toks_per_sent', default=65, type=int, help='Maximum number of tokens in a sentence')
  parser.add_argument('--concepts_per_sent', default=18, type=int, help='Maximum number of concepts in a sentence')
  parser.add_argument('--sentence_transformer_num_heads', default=4, type=int,
                      help='Number of attention heads in the sentence transformer')
  parser.add_argument('--sentence_transformer_num_stacks', default=2, type=int,
                      help='Number of times the sentence transformer is stacked')
  # rel
  parser.add_argument('--concept_mentions_per_doc', default=59, type=int,
                      help='Maximum number of concept mentions in a document')
  parser.add_argument('--concepts_per_doc', default=51, type=int,
                      help='Maximum number of unique concepts in a document')
  parser.add_argument('--mention_pairs', default=49, type=int, help='Maximum number of mention pairs in a document')
  parser.add_argument('--rel_hidden_size', default=100, type=int, help='Size of source/dest argument encodings')


  config = parser.parse_args(sys.argv[1:])

  config.record_dir = load_or_default(config.record_dir, config.base_data_dir.record_dir)
  config.json_data = load_or_default(config.json_data, config.base_data_dir.json_data)
  config.word2id = load_or_default(config.word2id, config.base_data_dir.word2id, loader=lambda x: json.load(open(x)))
  config.boundary2id = load_or_default(config.boundary2id, config.base_data_dir.boundary2id,
                                       loader=lambda x: json.load(open(x)))
  config.rel2id = load_or_default(config.rel2id, config.base_data_dir.rel2id, loader=lambda x: json.load(open(x)))

  run_switcher[config.mode](config)


if __name__ == "__main__":
  main()