Mercurial > ift6266

comparison baseline/mlp/mlp_get_error_from_model.py @ 212:e390b0454515

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
added classic lr time decay and py code to calculate the error based on a saved model
author xaviermuller
date Wed, 10 Mar 2010 16:17:59 -0500
parents
children 9b6e0af062af
comparison
equal deleted inserted replaced
211:476da2ba6a12 212:e390b0454515
1 __docformat__ = 'restructedtext en'
2
3 import pdb
4 import numpy as np
5 import pylab
6 import time
7 import pylearn
8 from pylearn.io import filetensor as ft
9
10 data_path = '/data/lisa/data/nist/by_class/'
11 test_data = 'all/all_test_data.ft'
12 test_labels = 'all/all_test_labels.ft'
13
14 def read_test_data(mlp_model):
15
16
17 #read the data
18 h = open(data_path+test_data)
19 i= open(data_path+test_labels)
20 raw_test_data = ft.read(h)
21 raw_test_labels = ft.read(i)
22 i.close()
23 h.close()
24
25 #read the model chosen
26 a=np.load(mlp_model)
27 W1=a['W1']
28 W2=a['W2']
29 b1=a['b1']
30 b2=a['b2']
31
32 return (W1,b1,W2,b2,raw_test_data,raw_test_labels)
33
34
35
36
37 def get_total_test_error(everything):
38
39 W1=everything[0]
40 b1=everything[1]
41 W2=everything[2]
42 b2=everything[3]
43 test_data=everything[4]/255.0
44 test_labels=everything[5]
45 total_error_count=0
46 total_exemple_count=0
47
48 nb_error_count=0
49 nb_exemple_count=0
50
51 char_error_count=0
52 char_exemple_count=0
53
54 min_error_count=0
55 min_exemple_count=0
56
57 maj_error_count=0
58 maj_exemple_count=0
59
60 for i in range(test_labels.size):
61 total_exemple_count = total_exemple_count +1
62 #get activation for layer 1
63 a0=np.dot(np.transpose(W1),np.transpose(test_data[i])) + b1
64 #add non linear function to layer 1 activation
65 a0_out=np.tanh(a0)
66
67 #get activation for output layer
68 a1= np.dot(np.transpose(W2),a0_out) + b2
69 #add non linear function for output activation (softmax)
70 a1_exp = np.exp(a1)
71 sum_a1=np.sum(a1_exp)
72 a1_out=a1_exp/sum_a1
73
74 predicted_class=np.argmax(a1_out)
75 wanted_class=test_labels[i]
76
77 if(predicted_class!=wanted_class):
78 total_error_count = total_error_count +1
79
80 #get grouped based error
81 if(wanted_class>9 and wanted_class<35):
82 min_exemple_count=min_exemple_count+1
83 predicted_class=np.argmax(a1_out)
84 if(predicted_class!=wanted_class):
85 min_error_count=min_error_count+1
86 elif(wanted_class<10):
87 nb_exemple_count=nb_exemple_count+1
88 predicted_class=np.argmax(a1_out)
89 if(predicted_class!=wanted_class):
90 nb_error_count=nb_error_count+1
91 elif(wanted_class>34):
92 maj_exemple_count=maj_exemple_count+1
93 predicted_class=np.argmax(a1_out)
94 if(predicted_class!=wanted_class):
95 maj_error_count=maj_error_count+1
96
97 if(wanted_class>9):
98 char_exemple_count=char_exemple_count+1
99 predicted_class=np.argmax(a1_out)
100 if(predicted_class!=wanted_class):
101 char_error_count=char_error_count+1
102
103
104 #convert to float
105 return ( total_exemple_count,nb_exemple_count,char_exemple_count,min_exemple_count,maj_exemple_count,\
106 total_error_count,nb_error_count,char_error_count,min_error_count,maj_error_count,\
107 total_error_count*100.0/total_exemple_count*1.0,\
108 nb_error_count*100.0/nb_exemple_count*1.0,\
109 char_error_count*100.0/char_exemple_count*1.0,\
110 min_error_count*100.0/min_exemple_count*1.0,\
111 maj_error_count*100.0/maj_exemple_count*1.0)
112
113
114
115
116
117
118
119
120
121
122
123
124