-
Notifications
You must be signed in to change notification settings - Fork 1
/
results_Tax2008_multiclass.py
348 lines (298 loc) · 15.3 KB
/
results_Tax2008_multiclass.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
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
from __future__ import division
import os
from optparse import OptionParser
import numpy as np
from sklearn.cross_validation import StratifiedKFold
from sklearn.svm import SVC
from sklearn.svm import OneClassSVM
from sklearn.mixture import GMM
from sklearn.neighbors import KernelDensity
from cwc.data_wrappers.datasets import Data
from cwc.models.discriminative_models import MyDecisionTreeClassifier
from cwc.models.background_check import BackgroundCheck
from cwc.models.oc_decomposition import OcDecomposition
from cwc.models.density_estimators import MyMultivariateKernelDensity
from cwc.models.confident_classifier import ConfidentClassifier
import pandas as pd
from diary import Diary
import copy
# Not to crop the output columns
pd.set_option('expand_frame_repr', False)
def separate_sets(x, y, test_fold_id, test_folds):
x_test = x[test_folds == test_fold_id, :]
y_test = y[test_folds == test_fold_id]
x_train = x[test_folds != test_fold_id, :]
y_train = y[test_folds != test_fold_id]
return [x_train, y_train, x_test, y_test]
def generate_outliers(x, y, variance_multiplier=4.0, outlier_proportion=0.5):
outlier_class = np.amax(y) + 1
n = np.alen(y)
n_o = int(np.around(outlier_proportion * n))
means = x.mean(axis=0)
covs = variance_multiplier*np.cov(x.T)
outliers = np.random.multivariate_normal(means, covs, n_o)
x_o = np.vstack((x, outliers))
y_o = np.append(y, np.ones(n_o) * outlier_class)
return x_o, y_o
class MyDataFrame(pd.DataFrame):
def append_rows(self, rows):
dfaux = pd.DataFrame(rows, columns=self.columns)
return self.append(dfaux, ignore_index=True)
def export_datasets_description_to_latex(data, path='', index=True):
df_data = MyDataFrame(columns=['Name', 'Samples', 'Features', 'Classes'])
dataset_names = data.datasets.keys()
dataset_names.sort()
for name in dataset_names:
dataset = data.datasets[name]
df_data = df_data.append_rows([[name, dataset.data.shape[0],
dataset.data.shape[1], len(dataset.classes)]])
def float_to_int_string(x):
return '%1.0f' % x
df_data.index += 1
df_data.to_latex(os.path.join(path,'datasets.tex'),
float_format=float_to_int_string , index=index)
def export_summary(df, diary):
def float_100_to_string(x):
return '%2.2f' % (100*x)
df = df.convert_objects(convert_numeric=True)
table = df.pivot_table(values=['acc', 'logloss'], index=['dataset'],
columns=['method'], aggfunc=[np.mean])
table.to_latex(os.path.join(diary.path,'acc.tex'),
float_format=float_100_to_string)
table = df.pivot_table(values=['acc', 'logloss'], index=['dataset'],
columns=['method'], aggfunc=[np.mean,
np.std])
table.to_latex(os.path.join(diary.path,'acc_std.tex'),
float_format=float_100_to_string)
diary.add_entry('summary', [table])
def fit_estimators(base_estimator, X, y):
estimators = []
bcs = []
classes = np.unique(y)
n_classes = np.alen(classes)
for c_index in np.arange(n_classes):
c = copy.deepcopy(base_estimator)
c.fit(X[y == c_index])
estimators.append(c)
bc = BackgroundCheck(estimator=base_estimator)
bc.set_estimator(c, X[y == c_index])
bcs.append(bc)
return estimators, bcs
def main(dataset_names=None, estimator_type="kernel", mc_iterations=1,
n_folds=10, seed_num=42):
if dataset_names is None:
datasets_li2014 = ['abalone', 'balance-scale', 'credit-approval',
'dermatology', 'ecoli', 'german', 'heart-statlog', 'hepatitis',
'horse', 'ionosphere', 'lung-cancer', 'libras-movement',
'mushroom', 'diabetes', 'landsat-satellite', 'segment',
'spambase', 'wdbc', 'wpbc', 'yeast']
datasets_hempstalk2008 = ['diabetes', 'ecoli', 'glass',
'heart-statlog', 'ionosphere', 'iris', 'letter',
'mfeat-karhunen', 'mfeat-morphological', 'mfeat-zernike',
'optdigits', 'pendigits', 'sonar', 'vehicle', 'waveform-5000']
datasets_others = [ 'diabetes', 'ecoli', 'glass', 'heart-statlog',
'ionosphere', 'iris', 'letter', 'mfeat-karhunen',
'mfeat-morphological', 'mfeat-zernike', 'optdigits',
'pendigits', 'sonar', 'vehicle', 'waveform-5000',
'scene-classification', 'tic-tac', 'autos', 'car', 'cleveland',
'dermatology', 'flare', 'page-blocks', 'segment', 'shuttle',
'vowel', 'zoo', 'abalone', 'balance-scale', 'credit-approval',
'german', 'hepatitis', 'lung-cancer']
# Datasets that we can add but need to be reduced
datasets_to_add = ['MNIST']
dataset_names = list(set(datasets_li2014 + datasets_hempstalk2008 +
datasets_others))
bandwidths_o_norm = {'glass': 0.09, 'hepatitis': 0.105,
'ionosphere': 0.039, 'vowel': 0.075}
bandwidths_bc = {'glass': 0.09, 'hepatitis': 0.105,
'ionosphere': 0.039, 'vowel': 0.0145}
bandwidths_t_norm = {'glass': 0.336, 'hepatitis': 0.015,
'ionosphere': 0.0385, 'vowel': 0.0145}
tuned_mus = {'glass': [0.094, 0.095, 0.2, 0.0, 0.0, 0.1],
'vowel': [0.0, 0.0, 0.5, 0.5, 0.5, 0.0]}
tuned_ms = {'glass': [1.0, 1.0, 1.0, 1.0, 1.0, 1.0],
'vowel': [1.0, 1.0, 1.0, 1.0, 1.0, 1.0]}
bandwidth_o_norm = 0.05
bandwidth_t_norm = 0.05
bandwidth_bc = 0.05
# Diary to save the partial and final results
diary = Diary(name='results_Tax2008', path='results',
overwrite=False,
fig_format='svg')
# Hyperparameters for this experiment (folds, iterations, seed)
diary.add_notebook('parameters', verbose=True)
# Summary for each dataset
diary.add_notebook('datasets', verbose=False)
# Partial results for validation
diary.add_notebook('validation', verbose=True)
# Final results
diary.add_notebook('summary', verbose=True)
columns=['dataset', 'method', 'mc', 'test_fold', 'acc']
df = MyDataFrame(columns=columns)
diary.add_entry('parameters', ['seed', seed_num, 'mc_it', mc_iterations,
'n_folds', n_folds,
'estimator_type', estimator_type,
'bw_o', bandwidth_o_norm,
'bw_t', bandwidth_t_norm,
'bw_bc', bandwidth_bc])
data = Data(dataset_names=dataset_names)
for name, dataset in data.datasets.iteritems():
if name in ['letter','shuttle']:
dataset.reduce_number_instances(0.1)
export_datasets_description_to_latex(data, path=diary.path)
for i, (name, dataset) in enumerate(data.datasets.iteritems()):
np.random.seed(seed_num)
dataset.print_summary()
diary.add_entry('datasets', [dataset.__str__()])
# accuracies_tuned = np.zeros(mc_iterations * n_folds)
# if name in bandwidths_o_norm.keys():
# bandwidth_o_norm = bandwidths_o_norm[name]
# bandwidth_t_norm = bandwidths_t_norm[name]
# bandwidth_bc = bandwidths_bc[name]
# else:
# bandwidth_o_norm = np.mean(bandwidths_o_norm.values())
# bandwidth_t_norm = np.mean(bandwidths_t_norm.values())
# bandwidth_bc = np.mean(bandwidths_bc.values())
for mc in np.arange(mc_iterations):
skf = StratifiedKFold(dataset.target, n_folds=n_folds,
shuffle=True)
test_folds = skf.test_folds
for test_fold in np.arange(n_folds):
x_train, y_train, x_test, y_test = separate_sets(
dataset.data, dataset.target, test_fold, test_folds)
# if name in ['glass', 'hepatitis', 'ionosphere', 'thyroid',
# 'iris', 'heart-statlog', 'diabetes', 'abalone',
# 'mushroom', 'spambase']:
x_test, y_test = generate_outliers(x_test, y_test)
# elif name == 'vowel':
# x_train = x_train[y_train <= 5]
# y_train = y_train[y_train <= 5]
# y_test[y_test > 5] = 6
# elif dataset.n_classes > 2:
# x_train = x_train[y_train <= dataset.n_classes/2]
# y_train = y_train[y_train <= dataset.n_classes/2]
# y_test[y_test > dataset.n_classes/2] = dataset.n_classes+1
# else:
# continue
if estimator_type == "svm":
est = OneClassSVM(nu=0.5, gamma=1.0/x_train.shape[1])
elif estimator_type == "gmm":
est = GMM(n_components=1)
elif estimator_type == "gmm3":
est = GMM(n_components=3)
elif estimator_type == "kernel":
est = MyMultivariateKernelDensity(kernel='gaussian',
bandwidth=bandwidth_bc)
# estimators = None
# bcs = None
# if estimator_type == "kernel":
# estimators, bcs = fit_estimators(
# MyMultivariateKernelDensity(
# kernel='gaussian',
# bandwidth=bandwidth_bc),
# x_train, y_train)
#
# # Untuned background check
# bc = BackgroundCheck(estimator=est, mu=0.0, m=1.0)
# oc = OcDecomposition(base_estimator=bc)
# if estimators is None:
# oc.fit(x_train, y_train)
# else:
# oc.set_estimators(bcs, x_train, y_train)
# accuracy = oc.accuracy(x_test, y_test)
# diary.add_entry('validation', ['dataset', name,
# 'method', 'BC',
# 'mc', mc,
# 'test_fold', test_fold,
# 'acc', accuracy])
# df = df.append_rows([[name, 'BC', mc, test_fold, accuracy]])
#
# e = MyMultivariateKernelDensity(kernel='gaussian',
# bandwidth=bandwidth_o_norm)
# oc_o_norm = OcDecomposition(base_estimator=e,
# normalization="O-norm")
# if estimators is None:
# oc_o_norm.fit(x_train, y_train)
# else:
# oc_o_norm.set_estimators(estimators, x_train, y_train)
# accuracy_o_norm = oc_o_norm.accuracy(x_test, y_test)
# diary.add_entry('validation', ['dataset', name,
# 'method', 'O-norm',
# 'mc', mc,
# 'test_fold', test_fold,
# 'acc', accuracy_o_norm])
# df = df.append_rows([[name, 'O-norm', mc, test_fold,
# accuracy_o_norm]])
#
# e = MyMultivariateKernelDensity(kernel='gaussian',
# bandwidth=bandwidth_t_norm)
# oc_t_norm = OcDecomposition(base_estimator=e,
# normalization="T-norm")
# if estimators is None:
# oc_t_norm.fit(x_train, y_train)
# else:
# oc_t_norm.set_estimators(estimators, x_train, y_train)
# accuracy_t_norm = oc_t_norm.accuracy(x_test, y_test)
# diary.add_entry('validation', ['dataset', name,
# 'method', 'T-norm',
# 'mc', mc,
# 'test_fold', test_fold,
# 'acc', accuracy_t_norm])
# df = df.append_rows([[name, 'T-norm', mc, test_fold,
# accuracy_t_norm]])
# Tuned background check
# if name in tuned_mus.keys():
# mus = tuned_mus[name]
# ms = tuned_ms[name]
# else:
# mus = None
# ms = None
# bc = BackgroundCheck(estimator=est, mu=0.0, m=1.0)
# oc_tuned = OcDecomposition(base_estimator=bc)
# oc_tuned.fit(x_train, y_train, mus=mus, ms=ms)
# accuracy_tuned = oc_tuned.accuracy(x_test, y_test, mus=mus,
# ms=ms)
# accuracies_tuned[mc * n_folds + test_fold] = accuracy_tuned
# diary.add_entry('validation', ['dataset', name,
# 'method', 'BC-tuned',
# 'mc', mc,
# 'test_fold', test_fold,
# 'acc', accuracy_tuned])
# df = df.append_rows([[name, 'BC-tuned', mc, test_fold,
# accuracy_tuned]])
conf = ConfidentClassifier(estimator=est)
conf.fit(x_train, y_train)
accuracy_conf = conf.accuracy(x_test, y_test)
diary.add_entry('validation', ['dataset', name,
'method', 'BC(C4.5)',
'mc', mc,
'test_fold', test_fold,
'acc', accuracy_conf])
df = df.append_rows([[name, 'BC(C4.5)', mc, test_fold,
accuracy_conf]])
export_summary(df, diary)
def parse_arguments():
parser = OptionParser()
parser.add_option("-d", "--dataset-names", dest="dataset_names",
default=None, help="list of dataset names coma separated")
parser.add_option("-e", "--estimator", dest="estimator_type",
default='kernel', type='string',
help="Estimator to use for the background check")
parser.add_option("-m", "--mc-iterations", dest="mc_iterations",
default=20, type=int,
help="Number of Monte Carlo iterations")
parser.add_option("-f", "--n-folds", dest="n_folds",
default=5, type=int,
help="Number of folds for the cross-validation")
parser.add_option("-s", "--seed-num", dest="seed_num",
default=42, type=int,
help="Seed number for the random number generator")
return parser.parse_args()
if __name__ == '__main__':
(options, args) = parse_arguments()
if options.dataset_names is not None:
dataset_names = options.dataset_names.split(',')
else:
dataset_names = None
main(dataset_names, options.estimator_type, options.mc_iterations,
options.n_folds, options.seed_num)