-
Notifications
You must be signed in to change notification settings - Fork 1
/
pred.py
241 lines (187 loc) · 8.88 KB
/
pred.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
from sklearn.model_selection import KFold
import random
from torch.utils.data import Dataset, DataLoader,TensorDataset,random_split,SubsetRandomSampler, ConcatDataset
import torch
import argparse
import configparser as ConfigParser
import ast
import numpy as np
from models import Classifier_BBB, Classifier_ConvBBB
from uncertainty import entropy_MI, overlapping, GMM_logits, calibration
import matplotlib.pyplot as plt
import csv
import mirabest
import torchvision.transforms as transforms
import torch.nn.functional as F
import torchvision
from galaxy_mnist import GalaxyMNIST, GalaxyMNISTHighrez
from pathlib import Path
from cata2data import CataData
from mightee import MighteeZoo
from utils import Path_Handler
def credible_interval(samples, credibility):
'''
calculate credible interval - equi-tailed interval instead of highest density interval
samples values and indices
'''
mean_samples = samples.mean()
sorted_samples = np.sort(samples)
lower_bound = 0.5 * (1 - credibility)
upper_bound = 0.5 * (1 + credibility)
index_lower = int(np.round(len(samples) * lower_bound))
index_upper = int(np.round(len(samples) * upper_bound))
return sorted_samples, index_lower, index_upper, mean_samples
def uncert_vi(model, test_data_uncert, device, T, burnin, reduction, path):
test_data = test_data_uncert
transform = transforms.Compose([transforms.ToTensor(),transforms.Normalize((0.0031 ,), (0.0350,))])
test_data1 = test_data_uncert
if(test_data_uncert == 'MBFRConfident'):
#confident test set
test_data = mirabest.MBFRConfident(path, train=False,
transform=transform, target_transform=None,
download=False)
test_data1 = mirabest.MBFRConfident(path, train=False,
transform=None, target_transform=None,
download=False)
#uncomment for test set
indices = np.arange(0, len(test_data), 1)
elif(test_data_uncert == 'MBFRUncertain'):
# uncertain
test_data = mirabest.MBFRUncertain(path, train=False,
transform=transform, target_transform=None,
download=False)
test_data1 = mirabest.MBFRUncertain(path, train=False,
transform=None, target_transform=None,
download=False)
data_type = 'MBFR_Uncert'
elif(test_data_uncert == 'MBHybrid'):
#hybrid
test_data = mirabest.MBHybrid(path, train=True,
transform=transform, target_transform=None,
download=False)
test_data1 = mirabest.MBHybrid(path, train=True,
transform=None, target_transform=None,
download=False)
data_type = 'MBHybrid'
elif(test_data_uncert == 'Galaxy_MNIST'):
transform = torchvision.transforms.Compose([ torchvision.transforms.ToTensor(),
torchvision.transforms.Resize((150,150), antialias = True),
torchvision.transforms.Grayscale(),
])
# 64 pixel images
train_dataset = GalaxyMNISTHighrez(
root='./dataGalaxyMNISTHighres',
download=True,
train=True, # by default, or set False for test set
transform = transform
)
test_dataset = GalaxyMNISTHighrez(
root='./dataGalaxyMNISTHighres',
download=True,
train=False, # by default, or set False for test set
transform = transform
)
gal_mnist_test_loader = torch.utils.data.DataLoader(test_dataset, batch_size=104, shuffle = False)
for i, (x_test_galmnist, y_test_galmnist) in enumerate(gal_mnist_test_loader):
x_test_galmnist, y_test_galmnist = x_test_galmnist.to(device), y_test_galmnist.to(device)
y_test_galmnist = torch.zeros(104).to(device)
if(i==0):
break
test_data = x_test_galmnist
elif(test_data_uncert == 'mightee'):
transform = torchvision.transforms.Compose(
[
torchvision.transforms.ToTensor(),
torchvision.transforms.Resize(150), # Rescale to adjust for resolution difference between MIGHTEE & RGZ - was 70
torchvision.transforms.Normalize((1.59965605788234e-05,), (0.0038063037602458706,)),
]
)
paths = Path_Handler()._dict()
set = 'certain'
data = MighteeZoo(path=paths["mightee"], transform=transform, set="certain")
test_loader = DataLoader(data, batch_size=len(data))
# for i, (x_test, y_test) in enumerate(test_loader):
# x_test, y_test = x_test.to(device), y_test.to(device)
test_data = data
else:
print("Test data for uncertainty quantification misspecified")
logit = True
indices = np.arange(0, len(test_data), 1)
num_batches_test = 1
error_all = []
entropy_all = []
mi_all = []
aleat_all =[]
avg_error_mean = []
loss_all = []
for index in (indices):
x = torch.unsqueeze(torch.tensor(test_data[index][0]),0)
y = torch.unsqueeze(torch.tensor(test_data[index][1]),0)
#print("target is",y)
target = y.detach().numpy().flatten()[0]
samples_iter = 200
#for a single datapoint
output_ = []
logits_=[]
y_test_all = []
prediction = []
errors = []
with torch.no_grad():
i = 1
model.eval()
for j in range(samples_iter):
x_test, y_test = x.to(device), y.to(device)
loss, pred, complexity_cost, likelihood_cost, conv_complexity, linear_complexity, logits = model.sample_elbo(x_test, y_test, 1, i, num_batches_test,samples_batch=len(y_test), T=T, burnin=burnin, reduction=reduction, logit= logit)
softmax = torch.exp(pred)
outputs = logits
output_.append(softmax.cpu().detach().numpy().flatten())
logits_.append(outputs.cpu().detach().numpy().flatten())
# prediction.append(pred.cpu().detach().numpy().flatten()[0])
y_test_all.append(y_test.cpu().detach().numpy().flatten()[0])
softmax = np.array(output_)#.cpu().detach().numpy())
y_logits = np.array(logits_)#.cpu().detach().numpy())
# print(softmax.shape)
sorted_softmax_values, lower_index, upper_index, mean_samples = credible_interval(softmax[:, 0].flatten(), 0.64)
print("90% credible interval for FRI class softmax", sorted_softmax_values[lower_index], sorted_softmax_values[upper_index])
sorted_softmax_values_fr1 = sorted_softmax_values[lower_index:upper_index]
sorted_softmax_values_fr2 = 1 - sorted_softmax_values[lower_index:upper_index]
softmax_mean = np.vstack((mean_samples, 1-mean_samples)).T
softmax_credible = np.vstack((sorted_softmax_values_fr1, sorted_softmax_values_fr2)).T
entropy, mutual_info, entropy_singlepass = entropy_MI(softmax_credible,
samples_iter= len(softmax_credible[:,0]))
pred = np.argmax(softmax_credible, axis = 1)
# print(y_test, pred)
y_test_all = np.tile(target, len(softmax_credible[:,0]))
errors = np.mean((pred != y_test_all).astype('uint8'))
pred_mean = np.argmax(softmax_mean, axis = 1)
error_mean = (pred_mean != target)*1
# print(error_mean)
# print(errors)
error_all.append(errors)
avg_error_mean.append(error_mean)
entropy_all.append(entropy/np.log(2))
mi_all.append(mutual_info/np.log(2))
aleat_all.append(entropy_singlepass/np.log(2))
# if(index == 1):
# break
n_bins = 8
path_out = './'
uce_pe = calibration(path_out, np.array(error_all), np.array(entropy_all), n_bins, x_label = 'predictive entropy')
# print("Predictive Entropy")
# print("uce = ", np.round(uce, 2))
uce_mi = calibration(path_out, np.array(error_all), np.array(mi_all), n_bins, x_label = 'mutual information')
# print("Mutual Information")
# print("uce = ", np.round(uce, 2))
uce_ae = calibration(path_out, np.array(error_all), np.array(aleat_all), n_bins, x_label = 'average entropy')
# print("Average Entropy")
# print("uce = ", np.round(uce, 2))
print("mean and std of error")
print(error_all)
print(np.mean(error_all)*100)
print(np.std(error_all))
print("Average of expected error")
print((np.array(avg_error_mean)).mean())
print((np.array(avg_error_mean)).std())
mean_expected_error = np.array(avg_error_mean).mean()
std_expected_error = np.array(avg_error_mean).std()
return mean_expected_error, std_expected_error, uce_pe, uce_mi, uce_ae