DeepLoader/deeploader/util/verification.py at master · whut09/DeepLoader · GitHub

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
# -*- coding:utf-8 -*-
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import os,sys
import math
import numpy as np
import argparse
import pickle

from .distance import get_distance

def find_threshold_sort(pos,neg):
    pos_list = sorted(pos, key=lambda x: x[0])
    neg_list = sorted(neg, key=lambda x: x[0], reverse=True)
    pos_count = len(pos_list)
    neg_count = len(neg_list)
    correct = 0
    threshold = 0
    #print('sort pos')
    #print(pos_list)
    #print('sort neg')
    #print(neg_list)
    for i in range(min(pos_count, neg_count)):
        if pos_list[i][0] > neg_list[i][0]:
            correct = i
            threshold = (pos_list[i][0] + neg_list[i][0])/2
            break
    #print("%d/%d" % (correct, pos_count))
    precision = (correct * 2.0) / (pos_count + neg_count)
    return precision, threshold


def get_accuracy(pos_list,neg_list,threshold):
    pos_count = len(pos_list)
    neg_count = len(neg_list)
    correct = 0
    for i in range(pos_count):
        if pos_list[i][0] < threshold:
            correct += 1

    for i in range(neg_count):
        if neg_list[i][0] > threshold:
            correct += 1
    precision = float(correct) / (pos_count + neg_count)
    return precision


def best_threshold(pos_list, neg_list, thrNum = 10000):
    ts = np.linspace(-1, 1, thrNum*2+1)
    best_acc = 0
    best_t = 0
    for t in ts:
        acc = get_accuracy(pos_list, neg_list, t)
        if acc > best_acc:
            best_acc = acc
            best_t = t
    return best_acc, best_t


def test_kfold(pos_list, neg_list, k = 10):
    fold_size = len(pos_list)//k
    sum_acc = 0
    sum_thresh = 0
    sum_n = 0
    accu_list = []
    for i in range(k):
        val_pos = []
        val_neg = []
        test_pos = []
        test_neg = []
        for j in range(len(pos_list)):
            fi = j//fold_size
            if fi != i:
                val_pos.append(pos_list[j])
                val_neg.append(neg_list[j])
            else:
                test_pos.append(pos_list[j])
                test_neg.append(neg_list[j])
        precision, threshold = find_threshold_sort(val_pos, val_neg)
        accuracy = get_accuracy(test_pos, test_neg, threshold)
        accu_list.append(accuracy)
        sum_acc += accuracy
        sum_thresh += threshold
        sum_n += 1
        # verbose
        print('precision:%.4f threshold:%f' % (accuracy, threshold))
    return sum_acc/sum_n, sum_thresh/sum_n, accu_list

def compute_distance(pos_list, neg_list, dist_type = 'L2'):
    '''
    [
      [feat1, feat2, ..],
        ...
      [feat1, feat2, ..]
    ]
    '''
    # distance measure
    if isinstance(dist_type, str):
        dist_func = get_distance(dist_type)
    else:
        dist_func = dist_type
    # get dist
    pos_dist = []
    for i in pos_list:
        dist = dist_func(i[0], i[1])
        pos_dist.append([dist])

    neg_dist = []
    for i in neg_list:
        dist = dist_func(i[0], i[1])
        neg_dist.append([dist])
    return pos_dist, neg_dist


def verification(pos_list, neg_list, dist_type = 'L2'):
    '''
    [
      [feat1, feat2, ..],
        ...
      [feat1, feat2, ..]
    ]
    '''
    pos_dist, neg_dist = compute_distance(pos_list, neg_list, dist_type)
    precision, threshold, accu_list = test_kfold(pos_dist, neg_dist)
    pos = sorted(pos_dist, key=lambda x: x[0])
    neg = sorted(neg_dist, key=lambda x: x[0], reverse=True)
    pos = [x[0] for x in pos]
    neg = [x[0] for x in neg]
    acc, std = np.mean(accu_list), np.std(accu_list)

    return acc, std, threshold, pos, neg, accu_list