/
evaluate_metrics.py
153 lines (127 loc) · 5.44 KB
/
evaluate_metrics.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
import numpy as np
import torch
from sklearn.metrics import ndcg_score
from sklearn.metrics import precision_score
def hamming_loss(y_GT, predict):
GT_size = np.sum(y_GT, axis=1)
predict_label = np.zeros(predict.shape, dtype=int)
sorted = predict.argsort()
temp = 0
for i in range(GT_size.shape[0]):
index=sorted[i][-GT_size[i]:][::-1]
predict_label[i][index]=1
temp = temp + np.sum(y_GT[i] ^ predict_label[i])
hmloss = temp/(y_GT.shape[0]*y_GT.shape[1])
return hmloss
def FR(y_GT, predict):
GT_size = np.sum(y_GT, axis=1)
predict_label = np.zeros(predict.shape, dtype=int)
sorted = predict.argsort()
temp = 0
for i in range(GT_size.shape[0]):
index = sorted[i][-GT_size[i]:][::-1]
predict_label[i][index] = 1
if np.sum(y_GT[i] ^ predict_label[i])>0:
temp = temp + 1
fr = temp / y_GT.shape[0]
return fr
def TP_index(y_targets, predict):
GT_size = np.sum(y_targets, axis=1)
predict_label = np.zeros(predict.shape, dtype=int)
sorted = predict.argsort()
tp_flag = False
for i in range(GT_size.shape[0]):
index = sorted[i][-GT_size[i]:][::-1]
predict_label[i][index] = 1
if np.sum(y_targets[i] ^ predict_label[i])==0:
tp_flag = True
return tp_flag, predict_label
def nontargeted_TP_index(y_GT, predict, kvalue):
predict_label = np.zeros(predict.shape, dtype=int)
sorted = predict.argsort()
flag = True
for i in range(predict.shape[0]):
index = sorted[i][-kvalue:][::-1]
# index = torch.flip(index, [0])
predict_label[i][index] = 1
for j in index:
if y_GT[i][j] == 1:
flag = False
return flag, predict_label
def specific_TP_index(y_GT, specific_GT, predict, GT_index_at_k, GT_sorted_at_k, GT_sort_origin_at_k, origin_GT_num):
predict_label = np.zeros(predict.shape, dtype=int)
# sorted = predict.argsort()
flag = True
dif = y_GT - specific_GT
for i in range(predict.shape[0]):
index = GT_index_at_k[i]
# index = sorted[i][-kvalue:][::-1]
if index.size < origin_GT_num or (GT_sorted_at_k < GT_sort_origin_at_k).any(): #not (GT_sorted_at_k == GT_sort_origin_at_k).all():
flag = False
break
predict_label[i][index] = 1
for j in index:
if specific_GT[i][j] == 1 or dif[i][j] == 0:
flag = False
break
return flag, predict_label
def delta_n(specific_GT, predict, predict_ori, k_value):
index_topk = torch.argsort(predict, descending=True)[:, :k_value]
index_topk_ori = torch.argsort(predict_ori, descending=True)[:, :k_value]
GT_num_topk = torch.sum(specific_GT[torch.arange(specific_GT.shape[0]).unsqueeze(1), index_topk])
GT_num_topk_ori = torch.sum(specific_GT[torch.arange(specific_GT.shape[0]).unsqueeze(1), index_topk_ori])
return GT_num_topk_ori - GT_num_topk
def label_match(y_GT, predict,k_value):
GT_size = np.sum(y_GT, axis=1)
sorted = predict.argsort()
for i in range(GT_size.shape[0]):
index = sorted[i][-k_value:][::-1]
for j in index:
if y_GT[j]==1:
return False
return True
def predict_top_k_labels(predict_values, kvalue):
labels = []
for i in range(predict_values.shape[0]):
a = predict_values[i].argsort()[-kvalue:][::-1]
labels.append(np.asarray(a))
return np.asarray(labels)
def topk_acc_metric(y_GT, predict, kvalue):
count = 0
GT_label_index_list = []
for i in range(y_GT.shape[0]):
GT_label_index_list.append(np.where(y_GT[i] == 1)[0])
top_k_predict_labels_list = predict_top_k_labels(predict, kvalue)
for i in range(top_k_predict_labels_list.shape[0]):
if kvalue > GT_label_index_list[i].shape[0]:
count = count + int(set(top_k_predict_labels_list[i]).issuperset(GT_label_index_list[i]))
else:
count = count + int(set(GT_label_index_list[i]).issuperset(top_k_predict_labels_list[i]))
return count/top_k_predict_labels_list.shape[0]
def precision_at_k(y_GT, predict, k):
y_GT = y_GT.clone().cpu().detach().numpy()
predict = predict.clone().cpu().detach().numpy()
top_k_idx = np.argpartition(predict, kth=-k)[:, -k:]
top_k_targets = y_GT[np.arange(y_GT.shape[0])[:, np.newaxis], top_k_idx]
top_k_scores = np.ones((y_GT.shape[0], k))
return precision_score(top_k_targets, top_k_scores, average='micro')
def precision_at_k_instance(targets, scores, top_k_scores, k):
top_k_idx = np.argpartition(scores, kth=-k)[-k:]
top_k_targets = targets[top_k_idx]
return precision_score(top_k_targets, top_k_scores[:k], average='micro')
def mAP_at_k(y_GT, predict, K):
y_GT = y_GT.clone().cpu().detach().numpy()
predict = predict.clone().cpu().detach().numpy()
top_k_idx = np.argsort(predict)[:, ::-1][:, :K]
top_k_scores = np.ones(K)
top_k_targets = y_GT[np.arange(y_GT.shape[0])[:, np.newaxis], top_k_idx]
p_at_k = np.array([[precision_at_k_instance(t, p, top_k_scores, k) for k in range(1, K + 1)] for t, p in zip(y_GT, predict)])
n_k = np.sum(y_GT, axis=1)
tmp_k = np.ones(y_GT.shape[0]) * K
tag = n_k <= tmp_k
n_k[~tag] = K
return np.average(np.sum(p_at_k * top_k_targets, axis=1) / n_k)
def NDCG_at_k(y_GT, predict, k):
y_GT = y_GT.clone().cpu().detach().numpy()
predict = predict.clone().cpu().detach().numpy()
return ndcg_score(y_true=y_GT, y_score=predict, k=k)