-
Notifications
You must be signed in to change notification settings - Fork 0
/
histogram_vs_platt_binning.py
247 lines (213 loc) · 9.88 KB
/
histogram_vs_platt_binning.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
import argparse
import calibrators
import numpy as np
import matplotlib.pyplot as plt
from matplotlib import rc
import utils
parser = argparse.ArgumentParser()
parser.add_argument('--logits_file', default='cifar_logits.dat', type=str,
help='Name of file to load logits, labels pair.')
parser.add_argument('--num_bin_selection', default=500, type=int,
help='Number of examples to use for Platt Scaling.')
parser.add_argument('--num_binning', default=500, type=int,
help='Number of examples to use for binning.')
def eval_top_calibration(probs, logits, labels, plugin=True):
correct = (utils.get_top_predictions(logits) == labels)
data = list(zip(probs, correct))
bins = utils.get_discrete_bins(probs)
binned_data = utils.bin(data, bins)
if plugin:
return utils.plugin_ce(binned_data) ** 2
else:
return utils.improved_unbiased_square_ce(binned_data)
# print([np.mean(np.array(l)[:, 0]) for l in binned_data])
# print([np.mean(np.array(l)[:, 1]) for l in binned_data])
# print(utils.plugin_ce(binned_data, power=1))
# def estimator(data):
# binned_data = utils.bin(data, bins)
# return utils.plugin_ce(binned_data, power=2)
# return utils.bootstrap_uncertainty(data, estimator, num_samples=num_samples)
def eval_top_mse(probs, logits, labels):
correct = (utils.get_top_predictions(logits) == labels)
return np.mean(np.square(probs - correct))
def eval_marginal_calibration(probs, logits, labels, plugin=True):
ces = [] # Compute the calibration error per class, then take the average.
k = logits.shape[1]
labels_one_hot = utils.get_labels_one_hot(np.array(labels), k)
for c in range(k):
probs_c = probs[:, c]
labels_c = labels_one_hot[:, c]
data_c = list(zip(probs_c, labels_c))
bins_c = utils.get_discrete_bins(probs_c)
binned_data_c = utils.bin(data_c, bins_c)
ce_c = utils.plugin_ce(binned_data_c) ** 2
ces.append(ce_c)
return np.mean(ces)
def eval_marginal_mse(probs, logits, labels):
assert probs.shape == logits.shape
return np.mean(np.square(probs - logits))
def compare_calibrators(logits, labels, num_calibration, num_bins, Calibrators,
eval_calibration, eval_mse, resample=True):
assert len(logits) == len(labels)
if resample:
indices = np.random.choice(list(range(len(logits))),
size=num_calibration, replace=True)
else:
indices = np.array(list(range(len(logits))))
np.random.shuffle(indices)
shuffled_logits = [logits[i] for i in indices]
shuffled_labels = [labels[i] for i in indices]
train_logits = shuffled_logits[:num_calibration]
train_labels = shuffled_labels[:num_calibration]
if resample:
eval_logits = logits
eval_labels = labels
else:
eval_logits = shuffled_logits[num_calibration:]
eval_labels = shuffled_labels[num_calibration:]
l2_ces = []
mses = []
for Calibrator in Calibrators:
calibrator = Calibrator(num_calibration, num_bins)
calibrator.train_calibration(train_logits, train_labels)
calibrated_probs = calibrator.calibrate(eval_logits)
mid = eval_calibration(calibrated_probs, eval_logits, eval_labels, plugin=resample)
mse = eval_mse(calibrated_probs, eval_logits, eval_labels)
l2_ces.append(mid)
mses.append(mse)
return l2_ces, mses
def average_calibration(logits, labels, num_calibration, num_bins, Calibrators,
eval_calibration, eval_mse, num_trials=100, resample=True):
l2_ces, mses = [], []
for _ in range(num_trials):
cur_l2_ces, cur_mses = compare_calibrators(
logits, labels, num_calibration, num_bins, Calibrators, eval_calibration, eval_mse,
resample=resample)
l2_ces.append(cur_l2_ces)
mses.append(cur_mses)
l2_ce_means = np.mean(l2_ces, axis=0)
l2_ce_stddevs = np.std(l2_ces, axis=0) / np.sqrt(num_trials)
mses = np.mean(mses, axis=0)
mse_stddevs = np.std(mses, axis=0) / np.sqrt(num_trials)
return l2_ce_means, l2_ce_stddevs, mses, mse_stddevs
def vary_bin_calibration(logits, labels, num_calibration, num_bins_list,
Calibrators, eval_calibration, eval_mse, num_trials=100, resample=True):
ce_list = []
stddev_list = []
mse_list = []
for num_bins in num_bins_list:
l2_ce_means, l2_ce_stddevs, mses, mse_stddevs = average_calibration(
logits, labels, num_calibration, num_bins, Calibrators,
eval_calibration, eval_mse, num_trials, resample=resample)
ce_list.append(l2_ce_means)
stddev_list.append(l2_ce_stddevs)
mse_list.append(mses)
return np.transpose(ce_list), np.transpose(stddev_list), np.transpose(mse_list)
def plot_ces(bins_list, l2_ces, l2_ce_stddevs):
font = {'family' : 'normal',
'size' : 20}
rc('font', **font)
# 90% confidence intervals.
error_bars_90 = 1.645 * l2_ce_stddevs
plt.errorbar(
bins_list, l2_ces[0], yerr=[error_bars_90[0], error_bars_90[0]],
barsabove=True, color='red', capsize=4, label='histogram')
plt.errorbar(
bins_list, l2_ces[1], yerr=[error_bars_90[1], error_bars_90[1]],
barsabove=True, color='blue', capsize=4, label='variance-reduced')
plt.ylabel("L2 Squared Calibration Error")
plt.xlabel("Number of Bins")
plt.legend(loc='lower right')
plt.show()
def plot_mse_ce_curve(bins_list, l2_ces, mses, xlim=None, ylim=None):
font = {'family' : 'normal',
'size' : 20}
rc('font', **font)
def get_pareto_points(data):
pareto_points = []
def dominated(p1, p2):
return p1[0] >= p2[0] and p1[1] >= p2[1]
for datum in data:
num_dominated = sum(map(lambda x: dominated(datum, x), data))
if num_dominated == 1:
pareto_points.append(datum)
return pareto_points
print(get_pareto_points(list(zip(l2_ces[0], mses[0], bins_list))))
print(get_pareto_points(list(zip(l2_ces[1], mses[1], bins_list))))
l2ces0, mses0 = zip(*get_pareto_points(list(zip(l2_ces[0], mses[0]))))
l2ces1, mses1 = zip(*get_pareto_points(list(zip(l2_ces[1], mses[1]))))
plt.title("MSE vs Calibration Error")
plt.scatter(l2ces0, mses0, c='red', marker='o', label='hist')
plt.scatter(l2ces1, mses1, c='blue', marker='s', label='ours')
plt.legend(loc='upper left')
if xlim is not None:
plt.xlim(xlim)
if ylim is not None:
plt.ylim(ylim)
plt.xlabel("L2 Squared Calibration Error")
plt.ylabel("Mean-Squared Error")
plt.show()
def cifar10_experiment_top_1_1_1000():
logits_file = 'cifar_logits.dat'
logits, labels = utils.load_test_logits_labels(logits_file)
bins_list = list(range(10, 101, 10))
num_trials = 100
num_calibration = 1000
l2_ces, l2_stddevs, mses = vary_bin_calibration(logits, labels, num_calibration,
bins_list,
Calibrators=[calibrators.HistogramTopCalibrator, calibrators.PlattBinnerTopCalibrator],
eval_calibration=eval_top_calibration, eval_mse=eval_top_mse, num_trials=num_trials,
resample=True)
plot_mse_ce_curve(bins_list, l2_ces, mses, xlim=(0.0, 0.002), ylim=(0.0425, 0.045))
plot_ces(bins_list, l2_ces, l2_stddevs)
def cifar10_experiment_top_1_2_3000():
logits_file = 'cifar_logits.dat'
logits, labels = utils.load_test_logits_labels(logits_file)
bins_list = list(range(10, 101, 10))
num_trials = 100
num_calibration = 3000
l2_ces, l2_stddevs, mses = vary_bin_calibration(logits, labels, num_calibration,
bins_list,
Calibrators=[calibrators.HistogramTopCalibrator, calibrators.PlattBinnerTopCalibrator],
eval_calibration=eval_top_calibration, eval_mse=eval_top_mse, num_trials=num_trials,
resample=True)
plot_mse_ce_curve(bins_list, l2_ces, mses, xlim=(0.0, 0.002), ylim=(0.0425, 0.045))
plot_ces(bins_list, l2_ces, l2_stddevs)
def cifar10_experiment_marginal_2_1_1000():
logits_file = 'cifar_logits.dat'
logits, labels = utils.load_test_logits_labels(logits_file)
bins_list = list(range(10, 101, 10))
num_trials = 100
num_calibration = 1000
l2_ces, l2_stddevs, mses = vary_bin_calibration(logits, labels, num_calibration,
bins_list,
Calibrators=[calibrators.HistogramMarginalCalibrator,
calibrators.PlattBinnerMarginalCalibrator],
eval_calibration=eval_marginal_calibration, eval_mse=eval_marginal_mse,
num_trials=num_trials, resample=True)
plot_mse_ce_curve(bins_list, l2_ces, mses, xlim=(0.0, 0.001), ylim=(0.0, 0.0075))
plot_ces(bins_list, l2_ces, l2_stddevs)
def cifar10_experiment_marginal_2_2_3000():
logits_file = 'cifar_logits.dat'
logits, labels = utils.load_test_logits_labels(logits_file)
bins_list = list(range(10, 101, 10))
num_trials = 20
num_calibration = 3000
l2_ces, l2_stddevs, mses = vary_bin_calibration(logits, labels, num_calibration,
bins_list,
Calibrators=[calibrators.HistogramMarginalCalibrator,
calibrators.PlattBinnerMarginalCalibrator],
eval_calibration=eval_marginal_calibration, eval_mse=eval_marginal_mse,
num_trials=num_trials, resample=True)
plot_mse_ce_curve(bins_list, l2_ces, mses, xlim=(0.0, 0.001), ylim=(0.0, 0.0075))
plot_ces(bins_list, l2_ces, l2_stddevs)
if __name__ == "__main__":
cifar10_experiment_top_1_1_1000()
# args = parser.parse_args()
# logits, labels = utils.load_test_logits_labels(args.logits_file)
# bins_list = list(range(5, 101, 5))
# l2_ces, l2_stddevs, mses = vary_bin_calibration(logits, labels, args.num_bin_selection,
# args.num_binning, bins_list, eval_top_calibration, eval_top_mse,
# [HistogramTopCalibrator, PlattBinnerTopCalibrator])
# plot_mse_ce_curve(bins_list, l2_ces, mses)
# plot_ces(bins_list, l2_ces, l2_stddevs)