-
Notifications
You must be signed in to change notification settings - Fork 4
/
Copy patheval_1nn.py
executable file
·169 lines (125 loc) · 6.89 KB
/
eval_1nn.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
import logging
import argparse
import json
from time import time
from datetime import datetime
from collections import defaultdict
from collections import Counter
import numpy as np
from nlm_encoder import TransformerEncoder
from vectorspace import VSM
from coarsewsd20_reader import coarse_senses
from coarsewsd20_reader import load_instances
from coarsewsd20_reader import ambiguous_words
from sklearn.metrics import f1_score, precision_score, recall_score
logging.basicConfig(level=logging.INFO,
format='%(asctime)s - %(levelname)s - %(message)s',
datefmt='%d-%b-%y %H:%M:%S')
def eval_nn(args):
all_sense_preds = defaultdict(list)
all_results = defaultdict(list)
# matching test instances
for amb_word in ambiguous_words:
logging.info('Evaluating %s ...' % amb_word)
for inst_idx, test_inst in enumerate(load_instances(amb_word, split='test', setname=args.dataset_id, mode=args.mode)):
gold_sense = test_inst['class']
if encoder.get_num_subtokens(test_inst['tokens']) >= args.max_seq_len:
logging.error('%s:%d exceeds max_seq_len (%d).' % (amb_word, inst_idx, args.max_seq_len))
preds = [('NULL', -1)]
all_sense_preds[gold_sense].append(preds)
all_results[amb_word].append((test_inst, preds))
continue
inst_vecs = encoder.token_embeddings([test_inst['tokens']])[0][0]
assert inst_vecs[test_inst['idx']][0] == amb_word # sanity check
amb_word_vec = inst_vecs[test_inst['idx']][1]
amb_word_vec = amb_word_vec / np.linalg.norm(amb_word_vec)
preds = senses_vsm.most_similar_vec(amb_word_vec, topn=None)
# filter preds for target word
preds = [(sense, score) for sense, score in preds if sense.split('_')[0] == amb_word]
all_sense_preds[gold_sense].append(preds)
all_results[amb_word].append((test_inst, preds))
# computing accuracies
all_senses_accs = {}
all_words_accs = {}
for amb_word in coarse_senses:
n_word_correct, n_word_insts = 0, 0
all_gold, all_pred = [], []
for sense in coarse_senses[amb_word]:
sense_preds = all_sense_preds[sense]
if len(sense_preds) == 0:
continue
n_sense_correct = sum([1 for preds in sense_preds if preds[0][0] == sense])
sense_acc = n_sense_correct / len(sense_preds)
all_senses_accs[sense] = sense_acc
n_word_correct += n_sense_correct
n_word_insts += len(sense_preds)
all_pred += [preds[0][0] for preds in sense_preds]
all_gold += [sense] * len(sense_preds)
word_recall_scores = recall_score(all_gold, all_pred, average=None)
word_recall_MFS = word_recall_scores[0]
word_recall_LFS = word_recall_scores[-1]
word_precision_scores = precision_score(all_gold, all_pred, average=None)
word_precision_MFS = word_precision_scores[0]
word_precision_LFS = word_precision_scores[-1]
print(amb_word, 'PRECISION', word_precision_MFS, word_precision_LFS)
print(amb_word, 'RECALL', word_recall_MFS, word_recall_LFS)
all_words_accs[amb_word] = n_word_correct / n_word_insts
# writing perf summary and logging to stdout
if args.mode != 'regular':
summary_path = 'results/%s/1nn/%s/summary.%s.csv' % (args.dataset_id, args.nlm_id, args.mode)
else:
summary_path = 'results/%s/1nn/%s/summary.csv' % (args.dataset_id, args.nlm_id)
with open(summary_path, 'w') as summary_f:
summary_f.write('word,sense,n_insts,acc\n')
for amb_word in coarse_senses:
n_word_insts = 0
for sense in coarse_senses[amb_word]:
if sense not in all_senses_accs:
continue
sense_acc = all_senses_accs[sense]
n_sense_insts = len(all_sense_preds[sense])
n_word_insts += n_sense_insts
summary_f.write('%s,%s,%d,%f\n' % (amb_word, sense, n_sense_insts, sense_acc))
word_acc = all_words_accs[amb_word]
summary_f.write('%s,%s,%d,%f\n' % (amb_word, 'ALL', n_word_insts, word_acc))
# store full results for further analysis
for amb_word in all_results:
if args.mode != 'regular':
word_results_path = 'results/%s/1nn/%s/%s.%s.jsonl' % (args.dataset_id, args.nlm_id, amb_word, args.mode)
else:
word_results_path = 'results/%s/1nn/%s/%s.jsonl' % (args.dataset_id, args.nlm_id, amb_word)
with open(word_results_path, 'w') as word_results_f:
for inst_idx, (test_inst, inst_matches) in enumerate(all_results[amb_word]):
jsonl_results = {'idx': inst_idx, 'matches': inst_matches, 'gold': test_inst['class'], 'tokens': test_inst['tokens']}
word_results_f.write('%s\n' % json.dumps(jsonl_results, sort_keys=True))
if __name__ == '__main__':
parser = argparse.ArgumentParser(description='Nearest Neighbors Evaluation.', formatter_class=argparse.ArgumentDefaultsHelpFormatter)
parser.add_argument('-nlm_id', help='HF Transfomers model name', required=False, default='bert-base-uncased')
parser.add_argument('-dataset_id', help='Dataset name', required=False, default='CoarseWSD-20')
parser.add_argument('-sv_path', help='Path to sense vectors', required=True)
parser.add_argument('-mode', type=str, default='regular', help='MFS or LFS', required=False,
choices=['regular', 'mfs', 'lfs'])
parser.add_argument('-max_seq_len', type=int, default=512, help='Maximum sequence length (BERT)', required=False)
parser.add_argument('-subword_op', type=str, default='mean', help='WordPiece Reconstruction Strategy', required=False,
choices=['mean', 'first', 'sum'])
parser.add_argument('-layers', type=str, default='-1 -2 -3 -4', help='Relevant NLM layers', required=False)
parser.add_argument('-layer_op', type=str, default='sum', help='Operation to combine layers', required=False,
choices=['mean', 'first', 'sum'])
args = parser.parse_args()
args.layers = [int(n) for n in args.layers.split(' ')]
if args.nlm_id not in args.sv_path.split('/')[-1].split('.'): # catch mismatched nlms/sense_vecs
logging.fatal("Provided sense vectors don't seem to match nlm_id (%s)." % args.nlm_id)
raise SystemExit('Fatal Error.')
encoder_cfg = {
'model_name_or_path': args.nlm_id,
'min_seq_len': 0,
'max_seq_len': args.max_seq_len,
'layers': args.layers,
'layer_op': 'sum',
'subword_op': 'mean'
}
logging.info('Loading NLM ...')
encoder = TransformerEncoder(encoder_cfg)
logging.info('Loading VSM ...')
senses_vsm = VSM(args.sv_path, normalize=True)
eval_nn(args)