-
Notifications
You must be signed in to change notification settings - Fork 2
/
Copy pathmetrics.py
327 lines (312 loc) · 13.4 KB
/
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
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
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
import math
import torch
import torch.nn as nn
import random
import numpy as np
import torch.nn.functional as F
import argparse
import os
import shutil
import time
import torch.nn.parallel
import torch.backends.cudnn as cudnn
import torch.optim
import torch.utils.data
import torchvision.transforms as transforms
import torchvision.datasets as datasets
from torch.autograd import Variable
from torch.utils.data import Dataset
from torch.utils.data.dataset import random_split
from torch.utils.data import DataLoader
import matplotlib.pyplot as plt
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
cifar_classes = ('plane', 'car', 'bird', 'cat',
'deer', 'dog', 'frog', 'horse', 'ship', 'truck')
def metrics_print_2step(net_mod, net_exp, expert_fn, n_classes, loader):
'''
Computes metrics for the confidence score method: on each example compare net_mod (classifier) and net_exp (expert) accuracies and defer
net_mod: classifier model
net_exp: expert error model (pytorch)
expert_fn: actual synth expert
n_classes: number of classes
loader: data loader
'''
correct = 0
correct_sys = 0
exp = 0
exp_total = 0
total = 0
real_total = 0
with torch.no_grad():
for data in loader:
images, labels, expert_preds, _, _ = data
images, labels, expert_preds = images.to(device), labels.to(device), expert_preds.to(device)
outputs_mod = net_mod(images)
outputs_exp = net_exp(images)
_, predicted = torch.max(outputs_mod.data, 1)
_, predicted_exp = torch.max(outputs_exp.data, 1)
batch_size = outputs_mod.size()[0] # batch_size
exp_prediction = expert_fn(images, labels)
for i in range(0, batch_size):
r_score = outputs_mod.data[i][predicted[i].item()].item()
r_score = outputs_exp.data[i][1].item() - r_score
r = 0
if r_score >= 0:
r = 1
else:
r = 0
if r == 0:
total += 1
correct += (predicted[i] == labels[i]).item()
correct_sys += (predicted[i] == labels[i]).item()
if r == 1:
exp += (exp_prediction[i] == labels[i].item())
correct_sys += (exp_prediction[i] == labels[i].item())
exp_total += 1
real_total += 1
cov = str(total) + str(" out of") + str(real_total)
to_print = {"coverage": cov, "system accuracy": 100 * correct_sys / real_total,
"expert accuracy": 100 * exp / (exp_total + 0.0002),
"classifier accuracy": 100 * correct / (total + 0.0001)}
return to_print
print(to_print)
def metrics_print_2step_linear(net_mod, net_exp, expert_fn, n_classes, loader):
'''
Computes metrics for the confidence score method with linear representations
net_mod: classifier model
net_exp: expert error model (pytorch)
expert_fn: actual synth expert
n_classes: number of classes
loader: data loader
'''
correct = 0
correct_sys = 0
exp = 0
exp_total = 0
total = 0
real_total = 0
with torch.no_grad():
for data in loader:
images, labels, expert_preds, _, images_orig = data
images, labels, expert_preds, images_orig = images.to(device), labels.to(device), expert_preds.to(device), images_orig.to(device)
outputs_mod = net_mod(images_orig)
outputs_exp = net_exp(images)
_, predicted = torch.max(outputs_mod.data, 1)
_, predicted_exp = torch.max(outputs_exp.data, 1)
batch_size = outputs_mod.size()[0] # batch_size
exp_prediction = expert_fn(images, labels)
for i in range(0, batch_size):
r_score = outputs_mod.data[i][predicted[i].item()].item()
r_score = outputs_exp.data[i][1].item() - r_score
r = 0
if r_score >= 0:
r = 1
else:
r = 0
if r == 0:
total += 1
correct += (predicted[i] == labels[i]).item()
correct_sys += (predicted[i] == labels[i]).item()
if r == 1:
exp += (exp_prediction[i] == labels[i].item())
correct_sys += (exp_prediction[i] == labels[i].item())
exp_total += 1
real_total += 1
cov = str(total) + str(" out of") + str(real_total)
to_print = {"coverage": cov, "system accuracy": 100 * correct_sys / real_total,
"expert accuracy": 100 * exp / (exp_total + 0.0002),
"classifier accuracy": 100 * correct / (total + 0.0001)}
return to_print
print(to_print)
def metrics_print(net, expert_fn, n_classes, loader):
'''
Computes metrics for deferal (L_{CE} loss method)
-----
Arguments:
net: model
expert_fn: expert model
n_classes: number of classes
loader: data loader
'''
correct = 0
correct_sys = 0
exp = 0
exp_total = 0
total = 0
real_total = 0
alone_correct = 0
correct_pred = {classname: 0 for classname in cifar_classes}
total_pred = {classname: 0 for classname in cifar_classes}
with torch.no_grad():
for data in loader:
images, labels, _, _ ,_ = data
images, labels = images.to(device), labels.to(device)
outputs = net(images)
_, predicted = torch.max(outputs.data, 1)
batch_size = outputs.size()[0] # batch_size
exp_prediction = expert_fn(images, labels)
for i in range(0, batch_size):
r = (predicted[i].item() == n_classes)
prediction = predicted[i]
final_pred = 0
if predicted[i] == n_classes:
max_idx = 0
# get second max
for j in range(0, n_classes):
if outputs.data[i][j] >= outputs.data[i][max_idx]:
max_idx = j
prediction = max_idx
else:
prediction = predicted[i]
alone_correct += (prediction == labels[i]).item()
if r == 0:
total += 1
final_pred = predicted[i]
correct += (predicted[i] == labels[i]).item()
correct_sys += (predicted[i] == labels[i]).item()
if r == 1:
final_pred = exp_prediction[i]
exp += (exp_prediction[i] == labels[i].item())
correct_sys += (exp_prediction[i] == labels[i].item())
exp_total += 1
real_total += 1
if labels[i].item() == final_pred:
correct_pred[cifar_classes[labels[i].item()]] += 1
total_pred[cifar_classes[labels[i].item()]] += 1
cov = str(total) + str(" out of") + str(real_total)
to_print = {"coverage": cov, "system accuracy": 100 * correct_sys / real_total,
"expert accuracy": 100 * exp / (exp_total + 0.0002),
"classifier accuracy": 100 * correct / (total + 0.0001),
"alone classifier": 100 * alone_correct / real_total}
print(to_print)
for classname, correct_count in correct_pred.items():
accuracy = 100 * float(correct_count) / total_pred[classname]
print("Accuracy for class {:5s} is: {:.3f} %".format(classname,
accuracy))
return to_print
def metrics_print_oracle(net_class, expert_fn, expert_k, n_classes, loader):
'''
Computes metrics for Oracle method (defer when expert is correct)
net_mod: classifier model
expert_fn: actual synth expert
expert_k: number of classes expert can predict
n_classes: number of classes
loader: data loader
'''
correct = 0
correct_sys = 0
exp = 0
exp_total = 0
total = 0
real_total = 0
correct_pred = {classname: 0 for classname in cifar_classes}
total_pred = {classname: 0 for classname in cifar_classes}
with torch.no_grad():
for data in loader:
images, labels, _, _ ,_ = data
images, labels = images.to(device), labels.to(device)
outputs_class = net_class(images)
_, predicted = torch.max(outputs_class.data, 1)
batch_size = outputs_class.size()[0] # batch_size
exp_prediction = expert_fn(images, labels)
for i in range(0, batch_size):
r = (expert_k >= labels[i].item())
final_pred = 0
#r = (exp_prediction[i] == labels[i].item()), this has noise
if r == 0:
total += 1
prediction = predicted[i]
if predicted[i] == n_classes:
max_idx = 0
for j in range(0, n_classes):
if outputs_class.data[i][j] >= outputs_class.data[i][max_idx]:
max_idx = j
prediction = max_idx
else:
prediction = predicted[i]
final_pred = prediction
correct += (prediction == labels[i]).item()
correct_sys += (prediction == labels[i]).item()
if r == 1:
final_pred = exp_prediction[i]
exp += (exp_prediction[i] == labels[i].item())
correct_sys += (exp_prediction[i] == labels[i].item())
exp_total += 1
real_total += 1
if labels[i].item() == final_pred:
correct_pred[cifar_classes[labels[i].item()]] += 1
total_pred[cifar_classes[labels[i].item()]] += 1
cov = str(total) + str(" out of") + str(real_total)
to_print = {"coverage": cov, "system accuracy": 100 * correct_sys / real_total,
"expert accuracy": 100 * exp / (exp_total + 0.0002),
"classifier accuracy": 100 * correct / (total + 0.0001)}
print(to_print)
for classname, correct_count in correct_pred.items():
accuracy = 100 * float(correct_count) / total_pred[classname]
print("Accuracy for class {:5s} is: {:.3f} %".format(classname,
accuracy))
def metrics_print_classifier(model, data_loader, defer_net = False):
'''
Prints metrics for classifier on label (no deferral)
model: model
data_loader: data loader
defer_net: boolean to indicate if model is a deferral module (has n_classes +1 outputs)
'''
# prepare to count predictions for each class
correct_pred = {classname: 0 for classname in cifar_classes}
total_pred = {classname: 0 for classname in cifar_classes}
correct = 0
total = 0
# again no gradients needed
with torch.no_grad():
for data in data_loader:
images, labels, _, _ ,_ = data
images, labels = images.to(device), labels.to(device)
outputs = model(images)
_, predictions = torch.max(outputs.data, 1) # maybe no .data
if defer_net:
predictions_fixed = predictions
for i in range(len(predictions_fixed)):
if predictions_fixed[i] == 10: #max class
max_idx = 0
# get second max
for j in range(0, 10):
if outputs.data[i][j] >= outputs.data[i][max_idx]:
max_idx = j
prediction = max_idx
predictions_fixed[i] = prediction
total += labels.size(0)
correct += (predictions == labels).sum().item()
# collect the correct predictions for each class
for label, prediction in zip(labels, predictions):
if label == prediction:
correct_pred[cifar_classes[label]] += 1
total_pred[cifar_classes[label]] += 1
print('Accuracy of the network on the %d test images: %.3f %%' % (len(data_loader),
100 * correct / total))
# print accuracy for each class
for classname, correct_count in correct_pred.items():
accuracy = 100 * float(correct_count) / total_pred[classname]
print("Accuracy for class {:5s} is: {:.3f} %".format(classname,
accuracy))
def metrics_print_expert(model, data_loader, defer_net = False):
'''
Computes metrics for expert model error prediction
model: model
data_loader: data loader
'''
correct = 0
total = 0
# again no gradients needed
with torch.no_grad():
for data in data_loader:
images, label, expert_pred, _ ,_ = data
expert_pred = expert_pred.long()
expert_pred = (expert_pred == label) *1
images, labels = images.to(device), expert_pred.to(device)
outputs = model(images)
_, predictions = torch.max(outputs.data, 1) # maybe no .data
total += labels.size(0)
correct += (predictions == labels).sum().item()
print('Accuracy of the network on the %d test images: %.3f %%' % (total,
100 * correct / total))