Mercurial > ift6266

comparison deep/stacked_dae/v_sylvain/nist_apriori_error.py @ 418:fb028b37ce92

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
Sert a calculer l'erreur sur les differentes classes de NIST.
author SylvainPL <sylvain.pannetier.lebeuf@umontreal.ca>
date Fri, 30 Apr 2010 14:47:25 -0400
parents
children 5ca2936f2062
comparison
equal deleted inserted replaced
417:0282882aa91f 418:fb028b37ce92
1 __docformat__ = 'restructedtext en'
2
3 import pdb
4 import numpy
5 from numpy import array
6 import time
7 import datetime
8 import pylearn
9 import copy
10 import sys
11 import os
12 import os.path
13 from pylearn.io import filetensor as ft
14 from jobman import DD
15 from ift6266 import datasets
16 import cPickle
17 from copy import copy
18 import math
19
20 from config import *
21
22 data_path = '/data/lisa/data/nist/by_class/'
23 test_data = 'all/all_train_data.ft'
24 test_labels = 'all/all_train_labels.ft'
25 state = DD(DEFAULT_HP_NIST)
26
27 #sda_model -> path for the parameters file
28 #dataset -> the dataset we use for the test
29 #part -> 0=train, 1=valid, 2=test
30 #type -> non-linearity type 0=sigmoid, 1=tanh
31 def test_data(sda_model,dataset,part=2,type=0):
32
33
34 f = open(sda_model)
35 parameters_pre=cPickle.load(f)
36 f.close()
37
38 W1 = array(copy(parameters_pre[0]))
39 #print 'W1: ' + str(W1.shape)
40 b1 = array(copy(parameters_pre[1]))
41 #print 'b1: ' + str(b1.shape)
42 W2 = array(copy(parameters_pre[2]))
43 #print 'W2: ' + str(W2.shape)
44 b2 = array(copy(parameters_pre[3]))
45 #print 'b2: ' + str(b2.shape)
46 W3 = array(copy(parameters_pre[4]))
47 #print 'W3: ' + str(W3.shape)
48 b3 = array(copy(parameters_pre[5]))
49 #print 'b3: ' + str(b3.shape)
50 if state['num_hidden_layers'] == 4:
51 W4 = array(copy(parameters_pre[6]))
52 b4 = array(copy(parameters_pre[7]))
53 Wo = array(copy(parameters_pre[8]))
54 bo = array(copy(parameters_pre[9]))
55 elif state['num_hidden_layers'] == 3:
56 Wo = array(copy(parameters_pre[6]))
57 #print 'Wo: ' + str(Wo.shape)
58 bo = array(copy(parameters_pre[7]))
59 #print 'bo: ' + str(bo.shape)
60 W4=None
61 b4=None
62 else:
63 print('Number of layers not implemented yet, please do it')
64
65
66 total_error_count=0
67 total_exemple_count=0
68 if part == 0:
69 iter = dataset.train(1)
70 if part == 1:
71 iter = dataset.valid(1)
72 if part == 2:
73 iter = dataset.test(1)
74 for x,y in iter:
75 total_exemple_count = total_exemple_count +1
76 if type == 1:
77 #get output for layer 1
78 out1=(numpy.tanh(numpy.dot(x,W1) + b1)+1.0)/2.0
79 #get output for layer 2
80 out2=(numpy.tanh(numpy.dot(out1,W2) + b2)+1.0)/2.0
81 #get output for layer 3
82 out3=(numpy.tanh(numpy.dot(out2,W3) + b3)+1.0)/2.0
83 #if there is a fourth layer
84 if state['num_hidden_layers'] == 4:
85 outf = (numpy.tanh(numpy.dot(out3,W4) + b4)+1.0)/2.0
86 else:
87 outf = array(out3)
88 else:
89 #get output for layer 1
90 out1=1.0/(1.0+numpy.exp(-(numpy.dot(x,W1)+b1)))
91 #get output for layer 2
92 out2 = 1.0/(1.0+numpy.exp(-(numpy.dot(out1,W2)+b2)))
93 #get output for layer 3
94 out3 = 1.0/(1.0+numpy.exp(-(numpy.dot(out2,W3)+b3)))
95 #if there is a fourth layer
96 if state['num_hidden_layers'] == 4:
97 outf = 1.0/(1.0+numpy.exp(-(numpy.dot(out3,W4)+b4)))
98 else:
99 outf = out3
100
101 out_act = numpy.dot(outf,Wo)+bo
102
103 #add non linear function for output activation (softmax)
104 #We can also use sigmoid and results will be the same
105 out = numpy.zeros(len(out_act[0]),float)
106 a1_exp = numpy.exp(out_act)
107 sum_a1=numpy.sum(a1_exp)
108 out=a1_exp/sum_a1
109 ## for i in xrange(len(out_act[0])):
110 ## out[i]=sigmoid(array(out_act[0,i]))
111
112 #get grouped based error
113 #with a priori
114 if(y>9 and y<35):
115 predicted_class=numpy.argmax(out[0,10:35])+10
116 if(predicted_class!=y):
117 total_error_count+=1
118
119 if(y<10):
120 predicted_class=numpy.argmax(out[0,0:10])
121 if(predicted_class!=y):
122 total_error_count+=1
123 if(y>34):
124 predicted_class=numpy.argmax(out[0,35:])+35
125 if(predicted_class!=y):
126 total_error_count+=1
127
128 print '\t total exemples count: '+str(total_exemple_count)
129 print '\t total error count: '+str(total_error_count)
130 print '\t percentage of error: '+str(total_error_count*100.0/total_exemple_count*1.0)+' %'
131
132
133 def sigmoid(value):
134 ## if len(value) > 1:
135 ## retour = numpy.zeros(len(value),float)
136 ## for i in xrange(len(value)):
137 ## retour[i] = (1.0/(1.0+math.exp(-float(value[i]))))
138 ## return retour
139 ## else:
140 ## print len(value)
141 return (1.0/(1.0+math.exp(-value)))
142
143 if __name__ == '__main__':
144
145 args = sys.argv[1:]
146
147 if len(args) > 0 and args[0] == 'sigmoid':
148 type = 0
149 elif len(args) > 0 and args[0] == 'tanh':
150 type = 1
151
152 part = 2 #0=train, 1=valid, 2=test
153
154 PATH = '' #Can be changed too if model is not in the current drectory
155
156 if os.path.exists(PATH+'params_finetune_NIST.txt'):
157 start_time = time.clock()
158 print ('\n finetune = NIST ')
159 print "NIST DIGITS"
160 test_data(PATH+'params_finetune_NIST.txt',datasets.nist_digits(),part=part,type=type)
161 print "NIST LOWER CASE"
162 test_data(PATH+'params_finetune_NIST.txt',datasets.nist_lower(),part=part,type=type)
163 print "NIST UPPER CASE"
164 test_data(PATH+'params_finetune_NIST.txt',datasets.nist_upper(),part=part,type=type)
165 end_time = time.clock()
166 print ('It took %f minutes' %((end_time-start_time)/60.))
167
168
169 if os.path.exists(PATH+'params_finetune_P07.txt'):
170 start_time = time.clock()
171 print ('\n finetune = P07 ')
172 print "NIST DIGITS"
173 test_data(PATH+'params_finetune_P07.txt',datasets.nist_digits(),part=part,type=type)
174 print "NIST LOWER CASE"
175 test_data(PATH+'params_finetune_P07.txt',datasets.nist_lower(),part=part,type=type)
176 print "NIST UPPER CASE"
177 test_data(PATH+'params_finetune_P07.txt',datasets.nist_upper(),part=part,type=type)
178 end_time = time.clock()
179 print ('It took %f minutes' %((end_time-start_time)/60.))
180
181
182 if os.path.exists(PATH+'params_finetune_NIST_then_P07.txt'):
183 start_time = time.clock()
184 print ('\n finetune = NIST then P07')
185 print "NIST DIGITS"
186 test_data(PATH+'params_finetune_NIST_then_P07.txt',datasets.nist_digits(),part=part,type=type)
187 print "NIST LOWER CASE"
188 test_data(PATH+'params_finetune_NIST_then_P07.txt',datasets.nist_lower(),part=part,type=type)
189 print "NIST UPPER CASE"
190 test_data(PATH+'params_finetune_NIST_then_P07.txt',datasets.nist_upper(),part=part,type=type)
191 end_time = time.clock()
192 print ('It took %f minutes' %((end_time-start_time)/60.))
193
194 if os.path.exists(PATH+'params_finetune_P07_then_NIST.txt'):
195 start_time = time.clock()
196 print ('\n finetune = P07 then NIST')
197 print "NIST DIGITS"
198 test_data(PATH+'params_finetune_P07_then_NIST.txt',datasets.nist_digits(),part=part,type=type)
199 print "NIST LOWER CASE"
200 test_data(PATH+'params_finetune_P07_then_NIST.txt',datasets.nist_lower(),part=part,type=type)
201 print "NIST UPPER CASE"
202 test_data(PATH+'params_finetune_P07_then_NIST.txt',datasets.nist_upper(),part=part,type=type)
203 end_time = time.clock()
204 print ('It took %f minutes' %((end_time-start_time)/60.))
205
206 if os.path.exists(PATH+'params_finetune_PNIST07.txt'):
207 start_time = time.clock()
208 print ('\n finetune = PNIST07')
209 print "NIST DIGITS"
210 test_data(PATH+'params_finetune_PNIST07.txt',datasets.nist_digits(),part=part,type=type)
211 print "NIST LOWER CASE"
212 test_data(PATH+'params_finetune_PNIST07.txt',datasets.nist_lower(),part=part,type=type)
213 print "NIST UPPER CASE"
214 test_data(PATH+'params_finetune_PNIST07.txt',datasets.nist_upper(),part=part,type=type)
215 end_time = time.clock()
216 print ('It took %f minutes' %((end_time-start_time)/60.))
217
218 if os.path.exists(PATH+'params_finetune_PNIST07_then_NIST.txt'):
219 start_time = time.clock()
220 print ('\n finetune = PNIST07 then NIST')
221 print "NIST DIGITS"
222 test_data(PATH+'params_finetune_PNIST07_then_NIST.txt',datasets.nist_digits(),part=part,type=type)
223 print "NIST LOWER CASE"
224 test_data(PATH+'params_finetune_PNIST07_then_NIST.txt',datasets.nist_lower(),part=part,type=type)
225 print "NIST UPPER CASE"
226 test_data(PATH+'params_finetune_PNIST07_then_NIST.txt',datasets.nist_upper(),part=part,type=type)
227 end_time = time.clock()
228 print ('It took %f minutes' %((end_time-start_time)/60.))
229
230
231
232
233
234
235
236
237
238
239