-
Notifications
You must be signed in to change notification settings - Fork 0
/
metric.py
125 lines (113 loc) · 5.51 KB
/
metric.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
import os
import pickle
import numpy as np
import pandas as pd
from sklearn.metrics import roc_curve, precision_recall_curve, auc, confusion_matrix, matthews_corrcoef, hamming_loss
from sklearn.metrics import accuracy_score, average_precision_score, precision_score,f1_score,recall_score
from util import *
def compute_auroc(labels, outputs):
fpr, tpr, _ = roc_curve(labels, outputs)
auc_val = auc(fpr, tpr)
return auc_val, fpr, tpr
def compute_auprc(labels, outputs):
precision, recall, _ = precision_recall_curve(labels, outputs)
auc_val = auc(recall, precision)
return auc_val, precision, recall
def compute_metric(labels, outputs, thres = 0.5, show_detail = True):
labels, outputs = np.array(labels), np.array(outputs)
tn, fp, fn, tp = confusion_matrix(labels, outputs > thres).ravel()
accuracy = (tp + tn) / (tn + fp + fn + tp)
sensitivity = tp / (tp + fp)
specificity = tn / (tn + fp)
precision = sensitivity
recall = tp / (tp+fn)
f1score = 2 * (precision * recall) / (precision + recall)
mcc = matthews_corrcoef(labels, outputs > thres)
fpr, tpr, _ = roc_curve(labels, outputs)
auc_val = auc(fpr, tpr)
if show_detail:
print(f'{round(accuracy, 4)}\t{round(sensitivity, 4)}\t{round(specificity, 4)}\t{round(mcc, 4)}\t{round(auc_val, 4)}')
return auc_val, fpr, tpr
def compute_metric_labelwise(labels, outputs, thres = 0.5, show_detail = True):
results_all = []
for i in range(len(all_labels)):
label = all_labels[i]
# print(label)
labels_i, outputs_i = labels[:, i], outputs[:, i]
tn, fp, fn, tp = confusion_matrix(labels_i, outputs_i > thres).ravel()
accuracy = (tp + tn) / (tn + fp + fn + tp + 1e-20)
sensitivity = tp / (tp + fp + 1e-20)
specificity = tn / (tn + fp + 1e-20)
precision = sensitivity
recall = tp / (tp+fn + + 1e-20)
f1score = 2 * (precision * recall) / (precision + recall + 1e-20)
mcc = matthews_corrcoef(labels_i, outputs_i > thres)
auc_val, precision, recall = compute_auroc(labels_i, outputs_i)
results_all.append([auc_val, accuracy, sensitivity, specificity, f1score, mcc])
if show_detail:
print(f'{round(auc_val, 3)}\t{round(accuracy,3)}\t{round(sensitivity,3)}\t{round(specificity,3)}\t{round(f1score,3)}\t{round(mcc, 3)}')
results_all = np.array(results_all)
metric_all = np.mean(results_all, axis = 0)
auc_val, accuracy, sensitivity, specificity, f1score, mcc = metric_all
print('Average')
print(f'{round(auc_val, 3)}\t{round(accuracy,3)}\t{round(sensitivity,3)}\t{round(specificity,3)}\t{round(f1score,3)}\t{round(mcc, 3)}')
def compute_metric_score(labels, outputs, thres = 0.5, show = True):
y_true, y_pred = labels, outputs > thres
'''
print('------Weighted------')
print('Weighted precision', precision_score(y_true, y_pred, average='weighted'))
print('Weighted recall', recall_score(y_true, y_pred, average='weighted'))
print('Weighted f1-score', f1_score(y_true, y_pred, average='weighted'))
'''
accuracy_per_class = [accuracy_score(y_true[:, i], y_pred[:, i]) for i in range(len(all_labels))]
# print(accuracy_per_class)
if show:
print('------Macro------')
# print(accuracy_per_class)
print('Macro accuracy', np.mean(accuracy_per_class))
print('Macro precision', precision_score(y_true, y_pred, average='macro'))
print('Macro recall', recall_score(y_true, y_pred, average='macro'))
print('Macro f1-score', f1_score(y_true, y_pred, average='macro'))
print('------Micro------')
print('Micro accuracy', accuracy_score(y_true.ravel(), y_pred.ravel()))
print('Micro precision', precision_score(y_true, y_pred, average='micro'))
print('Micro recall', recall_score(y_true, y_pred, average='micro'))
print('Micro f1-score', f1_score(y_true, y_pred, average='micro'))
print('------Hamming loss------')
print('Hamming loss', hamming_loss(y_true, y_pred))
return np.mean(accuracy_per_class)
def check_one_set(data_dir, pkl_path, show_iter = True, show_length = True, thres=0.5):
results = sorted(os.listdir(pkl_path))
num_split = len(results)
dataset = pd.read_csv(data_dir, sep='\t')
dataset['seqlen'] = dataset['text'].apply(lambda x: len(x))
for i in range(num_split):
pkl_dir = f'{pkl_path}/{results[i]}'
if not os.path.exists(pkl_dir):
print('Not existing', pkl_dir)
continue
with open(pkl_dir, 'rb') as fr:
pkl = pickle.load(fr)
outputs, labels = pkl
if show_iter:
compute_metric(labels[:, 1], outputs[:, 1], thres=thres)
dataset[i] = outputs[:, 1]
column_names = range(num_split)
averages = dataset[column_names].mean(axis=1)
print('Mean')
auroc_val, fpr, tpr = compute_metric(labels[:, 1], averages, thres=thres)
d_list = [dataset[(i*20 < dataset['seqlen']) & (dataset['seqlen'] <= (i + 1) * 20)] for i in range(3)]
for idx, d in enumerate(d_list):
if show_length:
print(f'Sequence Length from {idx * 20} to {(idx + 1) * 20}')
if (len(d) == 0):
continue
for i in range(num_split):
labels, outputs = d['label'].tolist(), d[i].tolist()
if show_iter and show_length:
compute_metric(labels, outputs)
averages = d[column_names].mean(axis=1)
if show_length:
print('Mean')
compute_metric(labels, averages)
return auroc_val, fpr, tpr