This repository has been archived by the owner on Apr 1, 2023. It is now read-only.
-
Notifications
You must be signed in to change notification settings - Fork 0
/
lexical_relation.py
223 lines (200 loc) · 8.29 KB
/
lexical_relation.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
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
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
import os
import logging
from glob import glob
import tqdm
from itertools import product
from multiprocessing import Pool
import pandas as pd
import numpy as np
from sklearn.metrics import f1_score
from sklearn.neural_network import MLPClassifier
from util import get_word_embedding_model, wget
logging.basicConfig(format='%(asctime)s %(levelname)-8s %(message)s', level=logging.INFO, datefmt='%Y-%m-%d %H:%M:%S')
pbar = tqdm.tqdm()
def get_lexical_relation_data():
""" get dataset """
cache_dir = 'cache'
os.makedirs(cache_dir, exist_ok=True)
root_url_analogy = 'https://github.com/asahi417/AnalogyTools/releases/download/0.0.0/lexical_relation_dataset.tar.gz'
if not os.path.exists('{}/lexical_relation_dataset'.format(cache_dir)):
wget(root_url_analogy, cache_dir)
full_data = {}
for i in glob('{}/lexical_relation_dataset/*'.format(cache_dir)):
if not os.path.isdir(i):
continue
full_data[os.path.basename(i)] = {}
label = {}
for t in glob('{}/*tsv'.format(i)):
with open(t) as f:
data = [line.split('\t') for line in f.read().split('\n') if len(line) > 0]
x = [d[:2] for d in data]
y = [d[-1] for d in data]
for _y in y:
if _y not in label:
label[_y] = len(label)
y = [label[_y] for _y in y]
full_data[os.path.basename(i)][os.path.basename(t).replace('.tsv', '')] = {'x': x, 'y': y}
full_data[os.path.basename(i)]['label'] = label
return full_data
def diff(a, b, model, add_feature='concat', pair_models=None, bi_direction: bool = True):
""" get feature for each pair """
try:
vec_a = model[a]
vec_b = model[b]
except KeyError:
return None
if 'concat' in add_feature:
feature = [vec_a, vec_b]
else:
feature = []
if 'diff' in add_feature:
feature.append(vec_a - vec_b)
if 'dot' in add_feature:
feature.append(vec_a * vec_b)
for pair_model in pair_models:
if pair_model is not None:
try:
vec_r = pair_model['__'.join([a, b]).lower().replace(' ', '_')]
except KeyError:
vec_r = np.zeros(pair_model.vector_size)
feature.append(vec_r)
if bi_direction:
try:
vec_r = pair_model['__'.join([b, a]).lower().replace(' ', '_')]
except KeyError:
vec_r = np.zeros(pair_model.vector_size)
feature.append(vec_r)
return np.concatenate(feature)
def run_test(clf, x, y):
""" run evaluation on valid or test set """
y_pred = clf.predict(x)
f_mac = f1_score(y, y_pred, average='macro')
f_mic = f1_score(y, y_pred, average='micro')
accuracy = sum([a == b for a, b in zip(y, y_pred.tolist())]) / len(y_pred)
return accuracy, f_mac, f_mic
class Evaluate:
def __init__(self, dataset, shared_config, default_config: bool = False):
self.dataset = dataset
if default_config:
self.configs = [{'random_state': 0}]
else:
learning_rate_init = [0.001, 0.0001, 0.00001]
# max_iter = [25, 50, 75]
hidden_layer_sizes = [100, 150, 200]
# learning_rate_init = [0.001, 0.0001]
# max_iter = [25]
# hidden_layer_sizes = [100]
self.configs = [{
'random_state': 0, 'learning_rate_init': i[0],
'hidden_layer_sizes': i[1]} for i in
list(product(learning_rate_init, hidden_layer_sizes))]
self.shared_config = shared_config
@property
def config_indices(self):
return list(range(len(self.configs)))
def __call__(self, config_id):
pbar.update(1)
config = self.configs[config_id]
# train
x, y = self.dataset['train']
clf = MLPClassifier(**config).fit(x, y)
# test
x, y = self.dataset['test']
t_accuracy, t_f_mac, t_f_mic = run_test(clf, x, y)
report = self.shared_config.copy()
report.update(
{'metric/test/accuracy': t_accuracy,
'metric/test/f1_macro': t_f_mac,
'metric/test/f1_micro': t_f_mic,
'classifier_config': clf.get_params()})
if 'val' in self.dataset:
x, y = self.dataset['val']
v_accuracy, v_f_mac, v_f_mic = run_test(clf, x, y)
report.update(
{'metric/val/accuracy': v_accuracy,
'metric/val/f1_macro': v_f_mac,
'metric/val/f1_micro': v_f_mic})
return report
def evaluate(embedding_model: str = None, feature='concat', add_relative: bool = False, add_pair2vec: bool = False):
model = get_word_embedding_model(embedding_model)
model_pair = []
if add_relative:
model_pair.append(get_word_embedding_model('relative_init.{}'.format(embedding_model)))
if add_pair2vec:
model_pair.append(get_word_embedding_model('pair2vec'))
data = get_lexical_relation_data()
report = []
for data_name, v in data.items():
logging.info('train model with {} on {}'.format(embedding_model, data_name))
label_dict = v.pop('label')
# preprocess data
oov = {}
dataset = {}
for _k, _v in v.items():
x = [diff(a, b, model, feature, model_pair) for (a, b) in _v['x']]
dim = len([_x for _x in x if _x is not None][0])
# initialize zero vector for OOV
dataset[_k] = [
[_x if _x is not None else np.zeros(dim) for _x in x],
_v['y']]
oov[_k] = sum([_x is None for _x in x])
shared_config = {
'model': embedding_model, 'feature': feature, 'add_relative': add_relative,
'add_pair2vec': add_pair2vec, 'label_size': len(label_dict), 'data': data_name,
'oov': oov
}
# grid serach
if 'val' not in dataset:
evaluator = Evaluate(dataset, shared_config, default_config=True)
tmp_report = evaluator(0)
else:
pool = Pool()
evaluator = Evaluate(dataset, shared_config)
tmp_report = pool.map(evaluator, evaluator.config_indices)
pool.close()
tmp_report = [tmp_report] if type(tmp_report) is not list else tmp_report
report += tmp_report
# print(report)
# print(pd.DataFrame(report))
# input()
del model
del model_pair
return report
if __name__ == '__main__':
# model_name = os.getenv('MODEL', 'w2v')
# print(model_name)
# target_word_embedding = [model_name]
target_word_embedding = ['w2v', 'fasttext', 'glove']
done_list = []
full_result = []
export = 'results/lexical_relation_all.csv'
if os.path.exists(export):
df = pd.read_csv(export, index_col=0)
done_list = df[['model', 'feature']].values.tolist()
full_result = [i.to_dict() for _, i in df.iterrows()]
logging.info("RUN WORD-EMBEDDING BASELINE")
pattern = ['diff', 'concat', ('diff', 'dot'), ('concat', 'dot')]
for m in target_word_embedding:
for _feature in pattern:
if [m, str(_feature)] in done_list:
continue
full_result += evaluate(m, feature=_feature)
if _feature in [('diff', 'dot'), ('concat', 'dot')]:
full_result += evaluate(m, feature=_feature, add_relative=True)
full_result += evaluate(m, feature=_feature, add_pair2vec=True)
pd.DataFrame(full_result).to_csv(export)
# aggregate result
# export = 'results/lexical_relation.{}.csv'.format(model_name)
export = 'results/lexical_relation.csv'
out = []
df = pd.DataFrame(full_result)
for _m in df.model.unique():
for _f in df.feature.unique():
for _d in df.data.unique():
for _r in df.add_relative.unique():
for _p in df.add_pair2vec.unique():
df_tmp = df[df.model == _m][df.feature == _f][df.data == _d][df.add_relative
== _r][df.add_pair2vec == _p]
df_tmp = df_tmp.sort_values(by=['metric/val/f1_macro'], ascending=False)
out.append(df_tmp.head(1))
pd.concat(out).to_csv(export)