-
Notifications
You must be signed in to change notification settings - Fork 0
/
ml_metric.py
195 lines (165 loc) · 7.68 KB
/
ml_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
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
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
import numpy as np
import torch
import matplotlib.pyplot as plt
from matplotlib.gridspec import GridSpec
import os
def Accuracy(target, pred, threshold=0.5):
# 首先把 y_pred 按照阈值给阈值化
pred[pred > threshold] = 1
pred[pred <= threshold] = 0
count = 0
for i in range(target.shape[0]):
p = sum(np.logical_and(target[i], pred[i]))
q = sum(np.logical_or(target[i], pred[i]))
count += p / q
return count / target.shape[0]
def F1Measure(target, pred, threshold=0.5):
# 首先把 pred 按照阈值给阈值化
pred[pred > threshold] = 1
pred[pred <= threshold] = 0
count = 0
for i in range(target.shape[0]):
if (sum(target[i]) == 0) and (sum(pred[i]) == 0):
continue
p = sum(np.logical_and(target[i], pred[i]))
q = sum(target[i]) + sum(pred[i])
count += (2 * p) / q
return count / target.shape[0]
def F1Measure_sklearn(target, pred, threshold=0.5):
# 首先把 pred 按照阈值给阈值化
pred[pred > threshold] = 1
pred[pred <= threshold] = 0
from sklearn.metrics import precision_score, recall_score, f1_score
# precision = precision_score(y_true=target, y_pred=pred, average='samples')
# recall = recall_score(y_true=target, y_pred=pred, average='samples')
f1measure = f1_score(y_true=target, y_pred=pred, average='samples')
return f1measure
def ECE_loss(target, pred, num_bin, threshold=0.5, network="", save_path="case_show"):
# target: (N, 20)
# pred: (N, 20)
# pred 是概率分布
bin_boundaries = torch.linspace(0, 1, num_bin + 1)
bin_lowers = bin_boundaries[:-1]
bin_uppers = bin_boundaries[1:]
# 计算 confidence
# 如果最大的 confidence > 0.5，Confidence 就是所有大于 0.5 的 confidence 做平均
# 如果最大的 confidence < 0.5，Confidence 就是最大的 confidence
confidences = []
for i in range(pred.shape[0]):
if pred[i].max().item() >= 0.5:
confidences.append(pred[i][pred[i] >= 0.5].mean().item())
else:
confidences.append(pred[i].max().item())
confidences = torch.from_numpy(np.array(confidences))
pred[pred > threshold] = 1
pred[pred <= threshold] = 0
count = 0
for i in range(target.shape[0]):
p = sum(np.logical_and(target[i], pred[i]))
q = sum(np.logical_or(target[i], pred[i]))
count += p / q
accuracy_overall = count / target.shape[0]
accuracies = []
for i in range(pred.shape[0]):
p = sum(torch.logical_and(target[i], pred[i]))
q = sum(torch.logical_or(target[i], pred[i]))
accuracies.append(p / q)
accuracies = torch.from_numpy(np.array(accuracies))
ece = torch.zeros(1, device=pred.device)
accuracy_bin = []
confidence_bin = []
count_bin = []
for bin_lower, bin_upper in zip(bin_lowers, bin_uppers):
# Calculated |confidence - accuracy| in each bin
in_bin = confidences.gt(bin_lower.item()) * confidences.le(bin_upper.item())
# prop_in_bin 是当前 bin 中样本数量占总样本数量的比例
prop_in_bin = in_bin.float().mean()
count_in_bin = in_bin.float().sum().item()
count_bin.append(count_in_bin)
# 如果 bin 中有样本则计算 ML-ECE
if prop_in_bin.item() > 0:
accuracy_in_bin = accuracies[in_bin].mean()
avg_confidence_in_bin = confidences[in_bin].mean()
accuracy_bin.append(accuracy_in_bin)
confidence_bin.append(confidence_bin)
ece += torch.abs(avg_confidence_in_bin - accuracy_in_bin) * prop_in_bin
else:
accuracy_bin.append(0)
confidence_bin.append(0)
import matplotlib.font_manager
plt.rcParams.update({
"text.usetex": True,
"font.family": "serif",
"font.serif": ["Times New Roman"],
'font.style': 'italic',
'font.weight': 'heavy', # or 'bold'
# 'font.size': 'medium',#or large,small
})
fig = plt.figure(figsize=(6, 9))
gs = GridSpec(3, 1)
ax1 = fig.add_subplot(gs[0, 0])
ax1.set_title(r'\textbf{%s}' % network, fontsize=20)
ax1.set_ylabel(r'\textbf{Sample frac}.', fontsize=20)
ax1.bar(bin_uppers, count_bin, width=1 / num_bin, color='#2776B7', edgecolor='white')
ax2 = fig.add_subplot(gs[1:, 0])
ax2.set_xlabel(r'\textbf{ML-Confidence (\%)}', fontsize=20)
ax2.set_ylabel(r'\textbf{ML-Accuracy (\%)}', fontsize=20)
ax2.plot([0.55, 1], [0.55, 1], color='gray', linewidth=4, linestyle="--")
ax2.bar(bin_uppers[10:], bin_uppers[10:], width=1 / num_bin, color='#2776B7', edgecolor='white',
label=r'\textbf{Gap}')
ax2.bar(bin_uppers, accuracy_bin, width=1 / num_bin, color='#7EB8DA', edgecolor='white', label=r'\textbf{Output}')
ax2.legend()
import matplotlib.patches as patches
ax2.add_patch(
patches.Rectangle(
(0.0, 0.13), # (x,y)
0.52, 0.2, # width and height
# You can add rotation as well with 'angle'
alpha=0.7, facecolor="white", edgecolor="lightgray", linewidth=3, linestyle='solid')
)
ax2.text(0.07, 0.15, r"$\mathbf {ECE_{ML}}$" + r"\textbf{ = %.3f}" % round(ece.item() * 100, 3) ,
color="black",
ha="left", va="bottom", transform=ax2.transAxes, fontsize=17)
ax2.text(0.07, 0.23, r"$\mathbf {ACC_{ML}}$" + r"\textbf{ = %.3f}" % round(accuracy_overall.item() * 100, 3),
color="black",
ha="left", va="bottom", transform=ax2.transAxes, fontsize=17)
plt.savefig(os.path.join(save_path, f'{network}_ML_ECE.png'), dpi=1500, format='png')
# plt.show()
plt.clf()
# fig, ax2 = plt.figure(figsize=(6, 6))
# ax2 = fig.add_subplot(gs[1:, 0])
fig = plt.figure(figsize=(6, 6))
gs = GridSpec(1, 1)
ax2 = fig.add_subplot(gs[0, 0])
ax2.set_title(r'\textbf{%s}' % network, fontsize=20)
ax2.set_xlabel(r'\textbf{ML-Confidence (\%)}', fontsize=20)
ax2.set_ylabel(r'\textbf{ML-Accuracy (\%)}', fontsize=20)
ax2.plot([0.55, 1], [0.55, 1], color='gray', linewidth=4, linestyle="--")
ax2.bar(bin_uppers[10:], bin_uppers[10:], width=1 / num_bin, color='#2776B7', edgecolor='white',
label=r'\textbf{Gap}')
ax2.bar(bin_uppers, accuracy_bin, width=1 / num_bin, color='#7EB8DA', edgecolor='white', label=r'\textbf{Output}')
ax2.legend()
import matplotlib.patches as patches
ax2.add_patch(
patches.Rectangle(
(0.0, 0.13), # (x,y)
0.52, 0.2, # width and height
# You can add rotation as well with 'angle'
alpha=0.7, facecolor="white", edgecolor="lightgray", linewidth=3, linestyle='solid')
)
# ax2.text(0.07, 0.21, r"{$\mathbf {ECE_{ML}}$}" + r"\textbf{ = %.3f}" % round(CPvVI, 3), color="black",
# ha="left", va="bottom", transform=ax2.transAxes, fontsize=18)
ax2.text(0.07, 0.15, r"$\mathbf {ECE_{ML}}$" + r"\textbf{ = %.3f }" % round(ece.item() * 100, 3) ,
color="black",
ha="left", va="bottom", transform=ax2.transAxes, fontsize=17)
ax2.text(0.07, 0.23, r"$\mathbf {ACC_{ML}}$" + r"\textbf{ = %.3f }" % round(accuracy_overall.item() * 100, 3),
color="black",
ha="left", va="bottom", transform=ax2.transAxes, fontsize=17)
plt.savefig(os.path.join(save_path, f'{network}_ML_ECE_lower_part.png'), dpi=1500, format='png')
plt.clf()
plt.title(f'{network}', fontsize=20)
plt.bar(bin_uppers, accuracy_bin, width=1 / num_bin, color='#7EB8DA', edgecolor='white', label='Output')
plt.xlabel('ML-Confidence', fontsize=20)
plt.ylabel('ML-Accuracy', fontsize=20)
plt.savefig(os.path.join(save_path, f'{network}_ML_ECE_true_calibrated.png'), dpi=1500, format='png')
return ece.item()