add evaluator
Browse files- evaluate_sample.py +25 -0
- evaluator.py +155 -0
evaluate_sample.py
ADDED
@@ -0,0 +1,25 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import json
|
2 |
+
from evaluator import prefix_accuracy, exact_match
|
3 |
+
|
4 |
+
if __name__ == '__main__':
|
5 |
+
path = './dataset/combined_dataset.json'
|
6 |
+
with open(path, 'r') as f:
|
7 |
+
data = [json.loads(line) for line in f]
|
8 |
+
|
9 |
+
fm_list, sm_list, pa_list = [], [], []
|
10 |
+
for item in data:
|
11 |
+
gt = item['label']
|
12 |
+
states_true = gt['intermediate'] + [gt['final']]
|
13 |
+
## dummy
|
14 |
+
states_pred = gt['intermediate'] + [gt['final']]
|
15 |
+
|
16 |
+
pa = prefix_accuracy(states_true, states_pred)
|
17 |
+
sm = pa == 1.0
|
18 |
+
fm = exact_match(states_true[-1], states_pred[-1])
|
19 |
+
pa_list.append(pa)
|
20 |
+
sm_list.append(sm)
|
21 |
+
fm_list.append(fm)
|
22 |
+
|
23 |
+
print(f'prefix accuracy: {sum(pa_list) / len(pa_list)}')
|
24 |
+
print(f'final match: {sum(fm_list) / len(fm_list)}')
|
25 |
+
print(f'sequential match: {sum(sm_list) / len(sm_list)}')
|
evaluator.py
ADDED
@@ -0,0 +1,155 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import numpy as np
|
2 |
+
|
3 |
+
def match_int(y_true, y_pred):
|
4 |
+
y_true_int = int(y_true)
|
5 |
+
try:
|
6 |
+
y_pred_int = int(y_pred)
|
7 |
+
except:
|
8 |
+
return 0.0
|
9 |
+
|
10 |
+
return float(y_true_int == y_pred_int)
|
11 |
+
|
12 |
+
def match_str(y_true, y_pred):
|
13 |
+
y_true_str = str(y_true)
|
14 |
+
try:
|
15 |
+
y_pred_str = str(y_pred)
|
16 |
+
except:
|
17 |
+
return 0.0
|
18 |
+
|
19 |
+
return float(y_true_str == y_pred_str)
|
20 |
+
|
21 |
+
def match_list(seq_true, seq_pred, match=match_str):
|
22 |
+
assert isinstance(seq_true, list)
|
23 |
+
|
24 |
+
if len(seq_true) == 0:
|
25 |
+
return 0.0
|
26 |
+
|
27 |
+
try:
|
28 |
+
seq_pred == list(seq_pred)
|
29 |
+
except:
|
30 |
+
return 0.0
|
31 |
+
|
32 |
+
if len(seq_true) != len(seq_pred):
|
33 |
+
return 0.0
|
34 |
+
|
35 |
+
arr = np.array([match(st, sp) for st, sp in zip(seq_true, seq_pred)])
|
36 |
+
return float(arr.all())
|
37 |
+
|
38 |
+
|
39 |
+
def reduce_last_intermediate(y_pred):
|
40 |
+
y_pred_ret = y_pred.copy()
|
41 |
+
if len(y_pred['intermediate']) == 0:
|
42 |
+
return y_pred_ret
|
43 |
+
if exact_match(y_pred['final'], y_pred['intermediate'][-1]):
|
44 |
+
y_pred_ret['intermediate'] = y_pred['intermediate'][:-1]
|
45 |
+
return y_pred_ret
|
46 |
+
|
47 |
+
def exact_match(y_true, y_pred):
|
48 |
+
type_y_true = type(y_true)
|
49 |
+
if type_y_true == str:
|
50 |
+
return match_str(y_true, y_pred)
|
51 |
+
elif type_y_true == int:
|
52 |
+
return match_int(y_true, y_pred)
|
53 |
+
elif type_y_true == list:
|
54 |
+
return match_list(y_true, y_pred)
|
55 |
+
else:
|
56 |
+
raise TypeError(f'type of ground truth is {type_y_true} and not supported')
|
57 |
+
|
58 |
+
def match_hit(seq_true, seq_pred):
|
59 |
+
assert type(seq_true) == list
|
60 |
+
if len(seq_true) == 0:
|
61 |
+
return np.nan
|
62 |
+
|
63 |
+
hit = 0.0
|
64 |
+
for st, sp in zip(seq_true, seq_pred):
|
65 |
+
hit += exact_match(st, sp)
|
66 |
+
return hit
|
67 |
+
|
68 |
+
def prefix_match_length(seq_true, seq_pred):
|
69 |
+
assert type(seq_true) == list
|
70 |
+
if len(seq_true) == 0:
|
71 |
+
return 0.0
|
72 |
+
|
73 |
+
hit = 0.0
|
74 |
+
for st, sp in zip(seq_true, seq_pred):
|
75 |
+
if exact_match(st, sp):
|
76 |
+
hit += 1
|
77 |
+
else:
|
78 |
+
break
|
79 |
+
return hit
|
80 |
+
|
81 |
+
def match_ratio(seq_true, seq_pred):
|
82 |
+
assert type(seq_true) == list
|
83 |
+
len_true = len(seq_true)
|
84 |
+
len_pred = len(seq_pred)
|
85 |
+
len_max = max(len_true, len_pred)
|
86 |
+
if len_true == 0:
|
87 |
+
return 0.0
|
88 |
+
|
89 |
+
hit = 0.0
|
90 |
+
for st, sp in zip(seq_true, seq_pred):
|
91 |
+
hit += exact_match(st, sp)
|
92 |
+
return hit / len_max
|
93 |
+
|
94 |
+
def prefix_accuracy(seq_true, seq_pred):
|
95 |
+
assert type(seq_true) == list
|
96 |
+
len_true = len(seq_true)
|
97 |
+
len_pred = len(seq_pred)
|
98 |
+
len_max = max(len_true, len_pred)
|
99 |
+
if len_true == 0:
|
100 |
+
return 0.0
|
101 |
+
|
102 |
+
hit = 0.0
|
103 |
+
for st, sp in zip(seq_true, seq_pred):
|
104 |
+
if exact_match(st, sp):
|
105 |
+
hit += 1
|
106 |
+
else:
|
107 |
+
break
|
108 |
+
return hit / len_max
|
109 |
+
|
110 |
+
|
111 |
+
if __name__ == '__main__':
|
112 |
+
assert match_int(5, '5') == 1.0
|
113 |
+
assert match_int(32, '32') == 1.0
|
114 |
+
assert match_int('512', 512) == 1.0
|
115 |
+
assert match_int(234, (234)) == 1.0
|
116 |
+
|
117 |
+
assert match_int(5, '55') == 0.0
|
118 |
+
assert match_int(512, 'a') == 0.0
|
119 |
+
assert match_int('512', 'a') == 0.0
|
120 |
+
assert match_int(231, [231]) == 0.0
|
121 |
+
assert match_int(231, [231, 66]) == 0.0
|
122 |
+
assert match_int(65, None) == 0.0
|
123 |
+
assert match_int(73, []) == 0.0
|
124 |
+
|
125 |
+
|
126 |
+
assert match_str('abc', 'abc') == 1.0
|
127 |
+
assert match_str('3', '3') == 1.0
|
128 |
+
assert match_str('21', 21) == 1.0
|
129 |
+
assert match_str('fe a t', 'fe a t') == 1.0
|
130 |
+
|
131 |
+
assert match_str('abc', 'ac') == 0.0
|
132 |
+
assert match_str('fe a t', 'feat') == 0.0
|
133 |
+
assert match_str('abc', ['abc']) == 0.0
|
134 |
+
assert match_str('abc', None) == 0.0
|
135 |
+
assert match_str('abc', []) == 0.0
|
136 |
+
|
137 |
+
|
138 |
+
assert match_list([1], [1]) == 1.0
|
139 |
+
assert match_list([1, 3, 5], [1, 3, 5]) == 1.0
|
140 |
+
assert match_list(['abc'], ['abc']) == 1.0
|
141 |
+
assert match_list(['abc', 'de'], ['abc', 'de']) == 1.0
|
142 |
+
|
143 |
+
assert match_list([], []) == 0.0
|
144 |
+
assert match_list([], [1]) == 0.0
|
145 |
+
assert match_list([2], [3]) == 0.0
|
146 |
+
assert match_list([1], [1, 1]) == 0.0
|
147 |
+
assert match_list([1], 1) == 0.0
|
148 |
+
assert match_list(['abc', 'de', 'f'], ['abc', 'de']) == 0.0
|
149 |
+
assert match_list(['abc', 'de'], ['abc', 'de', 'f']) == 0.0
|
150 |
+
assert match_list(['abc', 'de'], ['de', 'abc']) == 0.0
|
151 |
+
assert match_list([1], []) == 0.0
|
152 |
+
assert match_list([1], None) == 0.0
|
153 |
+
assert match_list(['abc', 'de'], [None, 'de']) == 0.0
|
154 |
+
|
155 |
+
print('All tests passed!')
|