/
test_chars74k.py
206 lines (161 loc) · 8.23 KB
/
test_chars74k.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
from __future__ import division
import numpy as np
from cwc.data_wrappers import reject
from cwc.evaluation.confidence_intervals import ConfidenceIntervals
from sklearn.cross_validation import StratifiedKFold
from sklearn.preprocessing import minmax_scale
from sklearn import svm
from sklearn.datasets import fetch_mldata
from sklearn import datasets
import matplotlib.pyplot as plt
from sklearn.cross_validation import train_test_split
from sklearn.grid_search import GridSearchCV
from sklearn.metrics import classification_report
from sklearn.preprocessing import label_binarize
def train_reject_model(x, r):
"""Train a classifier of training points
Returns a classifier that predicts high probability values for training
points and low probability values for reject points.
"""
model_rej = svm.SVC(C=1.0, gamma=0.016, kernel='rbf',
probability=True)
xr = np.vstack((x, r))
yr = np.hstack((np.ones(np.alen(x)), np.zeros(np.alen(r)))).T
model_rej.fit(xr, yr.astype(int))
return model_rej
def train_classifier_model(x, y):
model_clas = svm.SVC(C=10.0, gamma=0.002, kernel='rbf',
probability=True)
model_clas = model_clas.fit(x, y)
return model_clas
def calculate_mo_accuracy(model_clas_k, model_clas_u, model_rej_k,
model_rej_u, yk_test, yu_test):
yku_test = np.hstack((yk_test, yu_test)).reshape(-1, 1)
pred_clas_k = np.argmax(model_clas_k, axis=1)
pred_clas_u = np.argmax(model_clas_u, axis=1)
pred_clas = np.hstack((pred_clas_k, pred_clas_u)).reshape(-1,1)
y_rej = np.hstack((np.ones(np.alen(yk_test)), np.zeros(np.alen(yu_test)))).reshape(-1,1)
pred_rej_k = np.argmax(model_rej_k, axis=1)
pred_rej_u = np.argmax(model_rej_u, axis=1)
pred_rej = np.hstack((pred_rej_k, pred_rej_u)).reshape(-1,1)
multi_y = np.hstack((yku_test, y_rej))
multi_pred = np.hstack((pred_clas, pred_rej))
multi_pred_baseline = np.hstack((pred_clas, np.ones((np.alen(pred_rej),1))))
print('Accuracy f = {}'.format(np.mean(yk_test == pred_clas_k)))
print('Accuracy fku = {}'.format(np.mean(y_rej == pred_rej)))
multi_correct = multi_pred_baseline == multi_y
accuracy_baseline = np.sum(multi_correct) / (np.alen(y_rej) * 2)
print('Accuracy baseline = {}'.format(accuracy_baseline))
multi_correct = multi_pred == multi_y
accuracy_cco = np.sum(multi_correct) / (np.alen(y_rej) * 2)
print('Accuracy CCO = {}'.format(accuracy_cco))
return accuracy_baseline, accuracy_cco
def calculate_mo_ce(model_clas_k, model_clas_u, model_rej_k,
model_rej_u, yk_test, yu_test):
p_clas_k = label_binarize(yk_test, np.unique(yk_test))
p_clas_u = np.ones(p_clas_k.shape) / 10
p_clas = np.vstack((p_clas_k, p_clas_u))
q_clas = np.vstack((model_clas_k, model_clas_u))
e_clas = -np.sum(p_clas * np.log(np.clip(q_clas, 1e-16, 1.0)), axis=1)
p_rej = np.vstack([np.zeros((np.alen(yk_test), 1)), np.ones((np.alen(
yu_test),
1))])
p_rej = np.hstack([p_rej, 1-p_rej])
q_rej = np.vstack((model_rej_k, model_rej_u))
e_rej = -np.sum(p_rej * np.log(np.clip(q_rej, 1e-16, 1.0)), axis=1)
q_bas_rej = np.zeros(q_rej.shape)
q_bas_rej[:, 1] = 1.0
e_bas_rej = -np.sum(p_rej * np.log(np.clip(q_bas_rej, 1e-16, 1.0)), axis=1)
e_bas = np.sum(e_clas + e_bas_rej) / (np.alen(e_clas) * 2.0)
e_cco = np.sum(e_clas + e_rej) / (np.alen(e_clas) * 2.0)
print('Cross-entropy baseline = {}'.format(e_bas))
print('Cross-entropy CCO = {}'.format(e_cco))
return e_bas, e_cco
# best for f {'kernel': 'rbf', 'C': 10, 'gamma': 0.002}
# best for fku_hypershpere {'kernel': 'rbf', 'C': 1, 'gamma': 0.016}
# best for fku_hypercube {'kernel': 'rbf', 'C': 1, 'gamma': 0.001}
if __name__ == "__main__":
np.random.seed(1)
x = minmax_scale(np.load("datasets/chars74data.npy"), copy=False)
y = np.load("datasets/chars74target.npy")
digits_x = x[:550, :]
digits_y = y[:550]
letters_x = x[550:, :]
letters_y = y[550:]
mc_iterations = 100
accuracies_baseline = np.empty(mc_iterations)
accuracies_cco = np.empty(mc_iterations)
entropies_baseline = np.empty(mc_iterations)
entropies_cco = np.empty(mc_iterations)
for iteration in np.arange(mc_iterations):
print('Iteration = {}'.format(iteration+1))
X_train, xk_test, y_train, yk_test = train_test_split(digits_x, digits_y,
test_size=0.5,
random_state=0,
stratify=digits_y)
# Xu_train, xu_test, yu_train, yu_test = train_test_split(letters_x,
# letters_y,
# test_size=0.5,
# random_state=0)
nk_test = np.alen(yk_test)
nu_test = np.alen(letters_y)
chosen_indices = np.random.choice(nu_test, nk_test, replace=False)
xu_test = letters_x[chosen_indices, :]
yu_test = letters_y[chosen_indices]
model_clas = train_classifier_model(X_train, y_train)
u = reject.create_reject_data(X_train,
proportion=1, method='uniform_hsphere',
pca=True, pca_variance=0,
pca_components=10, hshape_cov=0,
hshape_prop_in=0.99, hshape_multiplier=2)
# u = np.random.uniform(0.0, 1.0, X_train.shape)
model_clas_ks = model_clas.predict_proba(X_train)
model_clas_us = model_clas.predict_proba(u)
p_x = np.hstack([model_clas_ks, X_train])
p_u = np.hstack([model_clas_us, u])
# xu = np.vstack((p_x, p_u))
# yu = np.hstack((np.ones(np.alen(p_x)), np.zeros(np.alen(p_u)))).T
#
# tuned_parameters = [{'kernel': ['rbf'], 'gamma': np.linspace(0.001, 0.1,
# 100),
# 'C': [1, 10, 100, 1000]},
# {'kernel': ['linear'], 'C': [1, 10, 100, 1000]}]
#
# print("# Tuning hyper-parameters for fku")
#
# clf = GridSearchCV(svm.SVC(C=1), tuned_parameters, cv=5,
# scoring='accuracy', verbose=1)
# clf.fit(xu, yu)
#
# print(clf.best_params_)
model_rej = train_reject_model(p_x, p_u)
model_clas_k = model_clas.predict_proba(xk_test)
model_clas_u = model_clas.predict_proba(xu_test)
p_xk_test = np.hstack([model_clas_k, xk_test])
p_xu_test = np.hstack([model_clas_u, xu_test])
model_rej_k = model_rej.predict_proba(p_xk_test)
model_rej_u = model_rej.predict_proba(p_xu_test)
# Multi-output accuracies
accuracies_baseline[iteration], accuracies_cco[iteration] = \
calculate_mo_accuracy(model_clas_k, model_clas_u, model_rej_k,
model_rej_u, yk_test, yu_test)
# Multi-output cross-entropies
entropies_baseline[iteration], entropies_cco[iteration] = \
calculate_mo_ce(model_clas_k, model_clas_u, model_rej_k,
model_rej_u, yk_test, yu_test)
# Confidence intervals for the multi-output accuracy
values = np.append(accuracies_baseline.reshape(-1, 1),
accuracies_cco.reshape(-1, 1), 1)
intervals = ConfidenceIntervals(values, ['Baseline', 'CCO'],
n_samples=100, alpha=0.05)
fig = plt.figure('accuracy')
plt.title('Multi-output Accuracy')
intervals.plot(fig)
# Confidence intervals for the multi-output cross-entropy
values = np.append(entropies_baseline.reshape(-1, 1),
entropies_cco.reshape(-1, 1), 1)
intervals = ConfidenceIntervals(values, ['Baseline', 'CCO'],
n_samples=100, alpha=0.05)
fig = plt.figure('cross-entropy')
plt.title('Multi-output Cross-entropy')
intervals.plot(fig)