-
Notifications
You must be signed in to change notification settings - Fork 1
/
cv_precomputed_scores.py
executable file
·202 lines (171 loc) · 6.52 KB
/
cv_precomputed_scores.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
#!/usr/bin/env python2.4
"""
Use cross validation to evaluate a model for some training data.
usage: %prog pos_data neg_data [options]
-F, --fold=N: Fold (default 5)
-M, --model=name: Name of model to train (default 'standard')
-l, --loo: Use leave-one-out cross validation (fold is ignored in this case)
"""
from __future__ import division
import array
import cookbook.doc_optparse
import sys
import traceback
import time
from itertools import *
from tempfile import mktemp
import commands
import math
import os
import random
import string
import sys
default_fold = 5
class CVClassification( object ):
def __init__( self ):
self.neg = 0
self.unc_neg = 0
self.unc_pos = 0
self.pos = 0
def get_total( self ):
return self.neg + self.unc_neg + self.unc_pos + self.pos
def __str__( self ):
return "%4d %4d %4d %4d" % ( self.pos, self.unc_pos, self.unc_neg, self.neg )
class CV( object ):
def __init__( self, data1, data2, fold, passes, loo ):
self.data1 = data1
self.data2 = data2
self.fold = fold
self.passes = passes
self.loo = loo
def get_success_rate( self ):
return ( float( self.cls1.unc_pos + self.cls1.pos + self.cls2.unc_neg + self.cls2.neg ) /
float( self.cls1.get_total() + self.cls2.get_total() ) )
def run( self ):
if self.loo: self.run_loo()
else: self.run_folds()
def run_folds( self ):
# Initialize classifications
self.cls1 = CVClassification()
self.cls2 = CVClassification()
# Run everything 'passes' times
for p in range( self.passes ):
# Create random partitions
partition1 = [ i % self.fold for i in range( len( self.data1 ) ) ]
random.shuffle( partition1 )
partition2 = [ i % self.fold for i in range( len( self.data2 ) ) ]
random.shuffle( partition2 )
# Run each fold
for f in range( self.fold ):
train1, test1 = self.split_by_partition( self.data1, partition1, f )
train2, test2 = self.split_by_partition( self.data2, partition2, f )
self.run_fold( train1, train2, test1, test2 )
def run_loo( self ):
# Initialize classifications
self.cls1 = CVClassification()
self.cls2 = CVClassification()
# Run everything 'passes' times
for p in range( self.passes ):
# Run for each item in positive set
for i in range( len( self.data1 ) ):
test = [ self.data1[i] ]
train = list( self.data1 )
del train[i]
self.run_fold( train, self.data2, test, [] )
# And each item in negative set
for i in range( len( self.data2 ) ):
test = [ self.data2[i] ]
train = list( self.data2 )
del train[i]
self.run_fold( self.data1, train, [], test )
def run_fold( self, train_set_1, train_set_2, test_set_1, test_set_2 ):
"""Run one fold of the cross validation"""
# Determine threshold
low, mid, high = self.determine_threshold( train_set_1, train_set_2 )
# Classify
self.classify( test_set_1, low, mid, high, self.cls1 )
self.classify( test_set_2, low, mid, high, self.cls2 )
def split_by_partition( self, set, partition, f ):
train, test = [], []
for i in range( len( set ) ):
if partition[i] == f: test.append( set[i] )
else: train.append( set[i] )
return train, test
def classify( self, scores, low, mid, high, cls ):
for score in scores:
if score < low: cls.neg += 1
elif score < mid: cls.unc_neg += 1
elif score < high: cls.unc_pos += 1
else: cls.pos += 1
def determine_threshold_simple( self, set1, set2 ):
smallest_pos = min( set1 )
largest_neg = max( set2 )
# If completely separated
if smallest_pos > largest_neg:
high = smallest_pos
low = largest_neg + 0.00000000001
mid = 0
# Else overlap
else:
high = low = mid = 0
# Return the thresholds
return low, mid, high
def determine_threshold( self, set1, set2 ):
sorted1 = set1[:]; sorted1.sort()
sorted2 = set2[:]; sorted2.sort()
# If completely separated
if sorted1[0] > sorted2[-1]:
high = sorted1[0]
low = sorted2[-1] + 0.00000000001
mid = ( high + low ) / 2.0
# Else overlap
else:
count1, count2 = len( set1 ), len( set2 )
index1, index2 = 0, 0
best_qual, best_score = 0.0, 0.0
while 1:
current_qual = ( ( float( count1 - index1 ) / float( count1 ) )
+ ( float( index2 ) / float( count2 ) ) )
if index2 < count2 and ( ( index1 == count1 ) or ( sorted2[ index2 ] < sorted1[ index1 ] ) ):
current_score = sorted2[ index2 ]
index2 += 1
elif index1 < count1:
current_score = sorted1[ index1 ]
index1 += 1
else:
break
if current_qual > best_qual:
best_score = current_score
best_qual = current_qual
high = low = mid = best_score
# Return the thresholds
return low, mid, high
def run( pos_file, neg_file, fold, loo ):
pos_strings = [ float( line ) for line in pos_file if line != "nan" ]
neg_strings = [ float( line ) for line in neg_file if line != "nan" ]
print "TP ~TP ~FN FN FP ~FP ~TN TN % time"
# Cross validate
if loo: passes = 1
else: passes = 5
cv_engine = CV( pos_strings, neg_strings, fold, passes, loo )
start_time = time.time()
cv_engine.run()
seconds = time.time() - start_time
print cv_engine.cls1, cv_engine.cls2,
print " %2.2f %2.2f" % ( cv_engine.get_success_rate()*100, seconds )
def main():
# Parse command line
options, args = cookbook.doc_optparse.parse( __doc__ )
#try:
if 1:
pos_fname, neg_fname = args
if options.fold:
fold = int( options.fold )
else:
fold = default_fold
loo = bool( options.loo )
#except:
# cookbook.doc_optparse.exit()
run( open( pos_fname ), open( neg_fname ), fold, loo )
if __name__ == "__main__":
main()