-
Notifications
You must be signed in to change notification settings - Fork 3
/
Copy pathclinical_eval.py
177 lines (151 loc) · 7.61 KB
/
clinical_eval.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
import copy
from collections import defaultdict
from data_objects import MultiheadConll
import argparse
def calculate_f1(tps, fps, fns):
p = 0. if not (tps + fps) else (tps / (tps + fps))
r = 0. if not (tps + fns) else (tps / (tps + fns))
f1 = 0. if not (p + r) else (2 * p * r / (p + r))
return p, r, f1
def evaluate_tuples(pred_tuples, gold_tuples, ix2rel, rel_col=-1):
# eval_dic[rel] = [tps, fps, fns]
tps_id = 0
fps_id = 1
fns_id = 2
eval_dic = defaultdict(lambda: [0, 0, 0])
for g_t in gold_tuples:
g_rel = ix2rel[g_t[rel_col]] if isinstance(g_t[rel_col], int) else g_t[rel_col]
if g_rel in ['N', 'O']:
continue
if g_t in pred_tuples:
eval_dic[g_rel][tps_id] += 1
pred_tuples.remove(g_t)
else:
eval_dic[g_rel][fns_id] += 1
for p_t in pred_tuples:
p_rel = ix2rel[p_t[rel_col]] if isinstance(p_t[rel_col], int) else p_t[rel_col]
if p_rel in ['N', 'O']:
continue
eval_dic[p_rel][fps_id] += 1
print()
for rel, (rel_tps, rel_fps, rel_fns) in eval_dic.items():
p, r, f1 = calculate_f1(rel_tps, rel_fps, rel_fns)
print("\t{:>12}, p {:.6f}, r {:.6f}, f1 {:.6f}, (tps {:d}, fps {:d}, fns {:d})".format(
rel,
p, r, f1,
rel_tps, rel_fps, rel_fns
))
all_tps = sum([v[tps_id] for v in eval_dic.values()])
all_fps = sum([v[fps_id] for v in eval_dic.values()])
all_fns = sum([v[fns_id] for v in eval_dic.values()])
all_p, all_r, all_f1 = calculate_f1(all_tps, all_fps, all_fns)
print("overall, p %.6f, r %.6f, f1 %.6f, (tps %i, fps %i, fns %i)\n" % (
all_p, all_r, all_f1,
all_tps, all_fps, all_fns
))
class TupleEvaluator(object):
# eval_dic[rel] = [tps, fps, fns]
def __init__(self):
self.tps_id = 0
self.fps_id = 1
self.fns_id = 2
self.eval_dic = defaultdict(lambda: [1e-10, 1e-10, 1e-10])
def reset(self):
self.eval_dic = defaultdict(lambda: [1e-10, 1e-10, 1e-10])
def update(self, gold_tuples, pred_tuples, rel_col=-1):
gold_tuple_cp = copy.deepcopy(gold_tuples)
pred_tuple_cp = copy.deepcopy(pred_tuples)
for g_t in gold_tuple_cp:
g_rel = g_t[rel_col]
if g_rel in ['N', 'O', '_', 'OO']:
continue
if g_t in pred_tuple_cp:
self.eval_dic[g_rel][self.tps_id] += 1
pred_tuple_cp.remove(g_t)
else:
self.eval_dic[g_rel][self.fns_id] += 1
for p_t in pred_tuple_cp:
p_rel = p_t[rel_col]
if p_rel in ['N', 'O', '_', 'OO']:
continue
self.eval_dic[p_rel][self.fps_id] += 1
def print_results(self, message, f1_mode, print_level):
class_scores = {}
for rel, (rel_tps, rel_fps, rel_fns) in self.eval_dic.items():
p, r, f1 = calculate_f1(rel_tps, rel_fps, rel_fns)
class_scores[rel] = (p, r, f1)
if print_level > 1:
print(f"\t{rel:>12}, p {p * 100:2.4f}, r {r * 100:2.4f}, f1 {f1 * 100:2.4f},"
f" (tps {rel_tps:.0f}, fps {rel_fps:.0f}, fns {rel_fns:.0f})")
if f1_mode == 'micro':
all_tps = sum([v[self.tps_id] for v in self.eval_dic.values()])
all_fps = sum([v[self.fps_id] for v in self.eval_dic.values()])
all_fns = sum([v[self.fns_id] for v in self.eval_dic.values()])
all_p, all_r, all_f1 = calculate_f1(all_tps, all_fps, all_fns)
elif f1_mode == 'macro':
all_p = sum([v[0] for k, v in class_scores.items()]) / len(class_scores)
all_r = sum([v[1] for k, v in class_scores.items()]) / len(class_scores)
all_f1 = sum([v[2] for k, v in class_scores.items()]) / len(class_scores)
else:
raise ValueError(f"Unknown f1_model: {f1_mode} ...")
if print_level >= 1:
print(f"{message}, {f1_mode} overall, p {all_p * 100:2.4f}, r {all_r * 100:2.4f}, f1 {all_f1 * 100:2.4f}")
return all_f1
class MhsEvaluator(object):
def __init__(self, gold_mhs_file, pred_mhs_file, f1_mode='micro'):
self._gold_mhs = MultiheadConll(gold_mhs_file)
self._pred_mhs = MultiheadConll(pred_mhs_file)
self.f1_mode = f1_mode
self._ner_evaluator = TupleEvaluator()
self._mod_evaluator = TupleEvaluator()
self._rel_evaluator = TupleEvaluator()
self._mention_rel_evaluator = TupleEvaluator()
# for g, p in zip(self._gold_mhs._rel_mention_triplets, self._pred_mhs._rel_mention_triplets):
# print(p)
# # print(g)
# print()
def eval_ner(self, print_level=1):
for s_gold_ner, s_pred_ner in zip(self._gold_mhs._entities, self._pred_mhs._entities):
self._ner_evaluator.update(s_gold_ner, s_pred_ner, rel_col=0)
return self._ner_evaluator.print_results('ner', f1_mode=self.f1_mode, print_level=print_level)
def eval_mod(self, print_level=1):
for s_gold_mod, s_pred_mod in zip(self._gold_mhs._mod_entities, self._pred_mhs._mod_entities):
self._mod_evaluator.update(s_gold_mod, s_pred_mod, rel_col=-1)
return self._mod_evaluator.print_results('mod', f1_mode=self.f1_mode, print_level=print_level)
def eval_rel_relax(self, print_level=1):
for s_gold_rel, s_pred_rel in zip(self._gold_mhs._rel_triplets, self._pred_mhs._rel_triplets):
self._rel_evaluator.update(s_gold_rel, s_pred_rel, rel_col=-1)
return self._rel_evaluator.print_results('rel (relax)', f1_mode=self.f1_mode, print_level=print_level)
def eval_rel(self, print_level=1):
for s_gold_rel, s_pred_rel in zip(self._gold_mhs._rel_detailed_triplets, self._pred_mhs._rel_detailed_triplets):
self._rel_evaluator.update(s_gold_rel, s_pred_rel, rel_col=-1)
return self._rel_evaluator.print_results('rel (strict)', f1_mode=self.f1_mode, print_level=print_level)
def eval_mention_rel(self, print_level=1):
for s_gold_rel, s_pred_rel in zip(self._gold_mhs._rel_mention_triplets, self._pred_mhs._rel_mention_triplets):
# print(s_pred_rel)
self._mention_rel_evaluator.update(s_gold_rel, s_pred_rel, rel_col=-1)
return self._mention_rel_evaluator.print_results('rel (str strict)', f1_mode=self.f1_mode, print_level=print_level)
if __name__ == '__main__':
parser = argparse.ArgumentParser(description='Clinical IE Evaluation')
parser.add_argument("--gold_file", default="data/i2b2/i2b2_test.conll", type=str,
help="gold file, multihead conll format.")
parser.add_argument("--pred_file", default="tmp/pred.conll", type=str,
help="pred file, multihead conll format.")
parser.add_argument("--eval_level", default=0, type=int,
help="0: all, 1: ner, 2: mod, 3: rel")
parser.add_argument("--print_level", default=2, type=int,
help="0: None, 1: F1scores, 2: detailed F1scores")
args = parser.parse_args()
evaluator = MhsEvaluator(args.gold_file, args.pred_file)
if args.eval_level == 0:
evaluator.eval_ner(print_level=args.print_level)
evaluator.eval_mod(print_level=args.print_level)
evaluator.eval_mention_rel(print_level=args.print_level)
evaluator.eval_rel(print_level=1)
elif args.eval_level == 1:
evaluator.eval_ner(print_level=args.print_level)
elif args.eval_level == 2:
evaluator.eval_mod(print_level=args.print_level)
elif args.eval_level == 3:
evaluator.eval_mention_rel(print_level=args.print_level)
evaluator.eval_rel(print_level=1)