-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathxmachina_kp.py
369 lines (274 loc) · 17.5 KB
/
xmachina_kp.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
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
# xmachina_kp.py
# Darren Temple
import csv
import numpy as np
import pickle
import random
import sys
from gramMatrix import gramMatrix
# --------------------------------------------------------------------------------------------------------------------------
# Variables
# --------------------------------------------------------------------------------------------------------------------------
# Oren trained on 225,000 samples, and tested on 25,000. 10,000 epochs. No normalisation.
verbose = True
superverbose = False
normalise = False # Works better with false
N_epoch = 5 # Default 100
eta = 0.00001 # Default 0.00001
find_best = False
#kernel_parameters_best = 4.5 # Cauchy
kernel_parameters_best = 44.82556976832949 # Gaussian
kernel_type = 'gaussian'
kernel_parameters = [i/10.0 for i in range(5, 51, 5)]
#kernel_type = 'exponential'
#kernel_parameters = [i/10.0 for i in range(5, 51, 5)]
#kernel_type = 'cauchy'
#kernel_parameters = [i/10.0 for i in range(5, 51, 5)]
#kernel_type = 'student'
#kernel_parameters = [i/10.0 for i in range(5, 51, 5)]
#kernel_type = 'power'
#kernel_parameters = [i/10.0 for i in range(5, 51, 5)]
#kernel_type = 'log'
#kernel_parameters = [i/10.0 for i in range(5, 51, 5)]
#kernel_type = 'sigmoid'
#kernel_parameters = [[5, 5]]
len_kernel_parameters = len(kernel_parameters)
# --------------------------------------------------------------------------------------------------------------------------
# Load the training data:
with open('../training_25000events.csv', 'rU') as IFile:
file_full = [row for row in csv.reader(IFile, delimiter = ',')]
features = file_full[0][1:-2]
D = len(features)
N = len(file_full) - 1
N_train = 20000
N_test = N - N_train
# Map from the given t = {'s', 'b'} to t = {+1, -1}:
data = [map(float, file_full[row][1:-2]) for row in range(1, N+1)]
data = np.array(data)
t = [ file_full[row][ -1] for row in range(1, N+1)]
t = [1 if i=='s' else -1 for i in t]
t = np.array(t)
del file_full
data_train = data[ 0:N_train]
data_test = data[N_train: ]
del data
t_train = t[ 0:N_train]
t_test = t[N_train: ]
del t
#for shuffle = 1:5
# randperm_N_train = randperm(N_train)
# data_train = data_train(randperm_N_train, :)
# t_train = t_train( randperm_N_train)
#end
if normalise:
mean_data_train = np.mean(data_train, axis = 0)
std_data_train = np.std (data_train, axis = 0)
data_train = np.subtract(data_train, mean_data_train)
data_train = np.divide (data_train, std_data_train)
data_test = np.subtract(data_test , mean_data_train)
data_test = np.divide (data_test , std_data_train)
# --------------------------------------------------------------------------------------------------------------------------
data_train_rowindex_all = range(N_train)
N_per_valset = 200 # Default 200
N_valset = N_train / N_per_valset
alpha = np.zeros(N_train)
if find_best:
N_incorrect_train_val = np.zeros((len_kernel_parameters, N_valset))
# --------------------------------------------------------------------------------------------------------------------------
# Calculation
# --------------------------------------------------------------------------------------------------------------------------
# Train kernel perceptron:
if find_best:
# Find the best kernel parameter(s):
if verbose:
print '\nFinding the best kernel parameter(s) ...\n'
for kernel_parameters_index in range(len_kernel_parameters):
if verbose:
print 'Parameter: %d/%d' % (kernel_parameters_index + 1, len_kernel_parameters)
K_train = gramMatrix(data_train, data_train,
kernel_type, kernel_parameters[kernel_parameters_index])
for valset in range(N_valset):
if verbose:
print ' Valset: %d/%d' % (valset + 1, N_valset)
data_train_rowindex_val = [i + valset * N_per_valset for i in range(N_per_valset)]
data_train_rowindex_use = range(valset * N_per_valset) + range((valset+1) * N_per_valset, N_train)
# Determine alpha for the current use set:
for epoch in range(N_epoch):
if verbose:
sys.stdout.write(' epoch: %d/%d\r' % (epoch + 1, N_epoch))
sys.stdout.flush()
data_train_use_rowindex_shuffle = range(N_train - N_per_valset)
random.shuffle(data_train_use_rowindex_shuffle)
for data_train_use_rowindex in data_train_use_rowindex_shuffle:
y_train_use_current = np.sign(np.dot( alpha[ data_train_rowindex_use],
K_train[np.ix_( data_train_rowindex_use,
[data_train_rowindex_use
[data_train_use_rowindex]] )] ))[0]
if y_train_use_current != t_train[data_train_rowindex_use[data_train_use_rowindex]]:
alpha[data_train_rowindex_use[data_train_use_rowindex]] \
= np.add( alpha[ data_train_rowindex_use
[data_train_use_rowindex] ],
np.multiply(eta, t_train[ data_train_rowindex_use
[data_train_use_rowindex] ]))
# end for data_train_use_rowindex
y_train_use = np.sign(np.dot( alpha[ data_train_rowindex_use],
K_train[np.ix_( data_train_rowindex_use,
data_train_rowindex_use )] ))
if 0 in y_train_use:
y_train_use[np.where(y_train_use == 0)]= 1
N_incorrect_train_use = sum(y_train_use != t_train[data_train_rowindex_use])
if superverbose:
print '\n N_incorrect_train_use: %d' % N_incorrect_train_use
elif verbose & (epoch == (N_epoch - 1)):
print ''
if verbose & (epoch < N_epoch-1) & (N_incorrect_train_use == 0):
print '\n N_incorrect_train_use = 0 =>break'
break
# end for epoch
# Try alpha with the current validation set:
y_train_val = np.sign(np.dot( alpha[ data_train_rowindex_use],
K_train[np.ix_( data_train_rowindex_use,
data_train_rowindex_val )] ))
if 0 in y_train_val:
y_train_val[np.where(y_train_val == 0)] = 1
N_incorrect_train_val[kernel_parameters_index, valset] \
= sum(y_train_val != t_train[data_train_rowindex_val])
alpha = np.zeros(N_train)
# end for valset
# end for kernel_parameters_index
sum_N_incorrect_train_val = np.sum(N_incorrect_train_val, axis = 1)
kernel_parameters_index_best = np.argmin(sum_N_incorrect_train_val)
kernel_parameters_best = kernel_parameters[kernel_parameters_index_best]
# --------------------------------------------------------------------------------------------------------------------------
# Retrain on the full dataset:
if verbose:
print '\nTraining ...\n'
if len_kernel_parameters > 1:
K_train = gramMatrix(data_train, data_train,
kernel_type, kernel_parameters_best)
for epoch in range(N_epoch):
if verbose:
sys.stdout.write(' epoch: %d/%d\r' % (epoch + 1, N_epoch))
sys.stdout.flush()
data_train_rowindex_shuffle = range(N_train)
random.shuffle(data_train_rowindex_shuffle)
for data_train_rowindex in data_train_rowindex_shuffle:
y_train_use_current = np.sign(np.dot(alpha, K_train[:,data_train_rowindex]))
if y_train_use_current != t_train[data_train_rowindex]:
alpha[data_train_rowindex] = np.add( alpha[data_train_rowindex],
np.multiply(eta, t_train[data_train_rowindex]))
# end for data_train_rowindex
y_train = np.sign(np.dot(alpha, K_train))
if 0 in y_train:
y_train[np.where(y_train == 0)] = 1
N_incorrect_train = sum(y_train != t_train)
if superverbose:
print '\n N_incorrect_train: %d' % N_incorrect_train
elif verbose & (epoch == (N_epoch - 1)):
print ''
if verbose & (epoch < N_epoch-1) & (N_incorrect_train == 0):
print '\n N_incorrect_train = 0 =>break'
break
# end for epoch
# --------------------------------------------------------------------------------------------------------------------------
# Test kernel perceptron:
K_test = gramMatrix(data_train, data_test,
kernel_type, kernel_parameters_best)
# Try alpha with the test set:
y_test = np.sign(np.dot(alpha, K_test))
if 0 in y_test:
y_test[np.where(y_test == 0)] = 1
N_incorrect_test = sum(y_test != t_test)
# Determine signal and background counts:
N_sig_t_train = sum(t_train == 1)
N_bkg_t_train = sum(t_train == -1)
N_sig_t_test = sum(t_test == 1)
N_bkg_t_test = sum(t_test == -1)
N_sig_y_train = sum(y_train == 1)
N_bkg_y_train = sum(y_train == -1)
N_sig_y_test = sum(y_test == 1)
N_bkg_y_test = sum(y_test == -1)
if (N_sig_y_train + N_bkg_y_train) != N_train:
print 'ERROR: (N_sig_y_train + N_bkg_y_train) ~= N_train'
if (N_sig_y_test + N_bkg_y_test ) != N_test :
print 'ERROR: (N_sig_y_test + N_bkg_y_test) ~= N_test'
# --------------------------------------------------------------------------------------------------------------------------
# Output
# --------------------------------------------------------------------------------------------------------------------------
if verbose:
print '\nOutput ...'
print ''
print ' Kernel type : %s' % kernel_type
print ' kernel parameters: %s' % str(kernel_parameters_best)
print ''
print ' Training: Total: %4d' % N_train
print ' N_sig: %4d (%6.2f%%) [Target: %4d (%6.2f%%)]' % \
(N_sig_y_train, float(N_sig_y_train) / float(N_train) * 100,
N_sig_t_train, float(N_sig_t_train) / float(N_train) * 100)
print ' N_bkg: %4d (%6.2f%%) [Target: %4d (%6.2f%%)]' % \
(N_bkg_y_train, float(N_bkg_y_train) / float(N_train) * 100,
N_bkg_t_train, float(N_bkg_t_train) / float(N_train) * 100)
print ' Misclassified: %4d (%6.2f%%)' % \
(N_incorrect_train, float(N_incorrect_train) / float(N_train) * 100)
print ''
print ' Testing : Total: %4d' % N_test
print ' N_sig: %4d (%6.2f%%) [Target: %4d (%6.2f%%)]' % \
(N_sig_y_test , float(N_sig_y_test ) / float(N_test ) * 100,
N_sig_t_test , float(N_sig_t_test ) / float(N_test ) * 100)
print ' N_bkg: %4d (%6.2f%%) [Target: %4d (%6.2f%%)]' % \
(N_bkg_y_test , float(N_bkg_y_test ) / float(N_test ) * 100,
N_bkg_t_test , float(N_bkg_t_test ) / float(N_test ) * 100)
print ' Misclassified: %4d (%6.2f%%)' % \
(N_incorrect_test , float(N_incorrect_test ) / float(N_test ) * 100)
print ''
# --------------------------------------------------------------------------------------------------------------------------
Pickle_Data = {}
Pickle_Data['normalise'] = normalise
Pickle_Data['N_epoch'] = N_epoch
Pickle_Data['eta'] = eta
Pickle_Data['find_best'] = find_best
Pickle_Data['kernel_type'] = kernel_type
Pickle_Data['kernel_parameters_best'] = kernel_parameters_best
Pickle_Data['D'] = D
Pickle_Data['N'] = N
Pickle_Data['N_trn'] = N_train
Pickle_Data['N_tst'] = N_test
if find_best:
Pickle_Data['N_valset'] = N_valset
Pickle_Data['N_per_valset'] = N_per_valset
Pickle_Data['t_trn'] = t_train
Pickle_Data['t_tst'] = t_test
Pickle_Data['y_trn'] = y_train
Pickle_Data['y_tst'] = y_test
Pickle_Data['N_sig_t_trn'] = N_sig_t_train
Pickle_Data['N_bkg_t_trn'] = N_bkg_t_train
Pickle_Data['N_sig_t_tst'] = N_sig_t_test
Pickle_Data['N_bkg_t_tst'] = N_bkg_t_test
Pickle_Data['N_sig_y_trn'] = N_sig_y_train
Pickle_Data['N_bkg_y_trn'] = N_bkg_y_train
Pickle_Data['N_sig_y_tst'] = N_sig_y_test
Pickle_Data['N_bkg_y_tst'] = N_bkg_y_test
if find_best:
Pickle_Data['N_incorrect_trn_val'] = N_incorrect_train_val
Pickle_Data['N_incorrect_trn'] = N_incorrect_train
Pickle_Data['N_incorrect_tst'] = N_incorrect_test
Pickle_File = open( 'xmachina_kp'
+ '_' + str(N)
+ '_' + str(N_epoch)
+ '_' + str(eta)
+ '_' + kernel_type
+ normalise * '_normalise'
+ '.pickle', 'w' )
pickle.dump(Pickle_Data, Pickle_File)
Pickle_File.close()
# To load the pickle and put the variables straight into memory ...
#
# import pickle
#
# Pickle_File = open('<pickle_file_name>', 'r')
# Pickle_Data = pickle.load(Pickle_File)
# locals().update(Pickle_Data)
#
# ... or more efficiently doing all three steps at once ...
#
# locals().update(pickle.load(open('<pickle_file_name>', 'r')))