/
openmax.py
182 lines (152 loc) · 7.09 KB
/
openmax.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
import sys
sys.path.insert(0, '../Tools')
import evaluation_tools as tools
import pandas as pd
import EVM
import libmr
import scipy
import functools
import cPickle
import numpy as np
from multiprocessing import Process,Pool,pool
import argparse
parser = argparse.ArgumentParser()
parser.add_argument("--rerun", dest="rerun", action="store_true", default=False)
parser.add_argument("--run_on_logits", dest="run_on_logits", action="store_true", default=False)
parser.add_argument("--use_euclidean", dest="use_euclidean", action="store_true", default=False)
parser.add_argument("--DIR_file_name", dest="dir_file_name", action="store", default=None)#, required=True)
parser.add_argument("--tail_size", type=int, dest="tail_size", action="store", default=15)
args = parser.parse_args()
alpha=5
def cls_mav_dist(cls):
data={}
same_cls = feature_vector_file.iloc[results_file.index[results_file[1]==cls],1:]
print "same_cls",same_cls.shape
MAV=same_cls.mean()
if args.use_euclidean:
distances=same_cls.apply(lambda row:scipy.spatial.distance.euclidean(MAV, row),axis=1)
else:
distances=same_cls.apply(lambda row:scipy.spatial.distance.cosine(MAV, row),axis=1)
distances=distances.values.tolist()
print "len(distances)",len(distances)
mr = libmr.MR()
# Fitting an EVT on the distances
mr.fit_high(distances,args.tail_size)
data['model']=mr
data['MAV']=MAV
data['distances']=distances
del distances,same_cls
return data
#if args.rerun:
if True:
feature_vector_file = pd.read_csv('Train_MNIST_openMax_deep_features.txt',
delimiter=' ',header=None)#,memory_map=True)#, nrows=10000)
results_file = pd.read_csv('Train_MNIST_openMax_pred.txt',
delimiter=' ',header=None)#,memory_map=True)#, nrows=10000)
"""
feature_vector_file = pd.read_csv('OpenMax/train_deep_features.txt',
delimiter=' ',header=None)#,memory_map=True)#, nrows=10000)
results_file = pd.read_csv('OpenMax/train_pred.txt',
delimiter=' ',header=None)#,memory_map=True)#, nrows=10000)
"""
print feature_vector_file.shape,results_file.shape
print feature_vector_file,results_file
# feature_vector_file = pd.read_csv('/home/adhamija/faceness/ImageNet/Pretrained_model/features_train_only_knowns.txt',
# delimiter=' ',header=None)#,memory_map=True)#, nrows=10000)
# results_file = pd.read_csv('/home/adhamija/faceness/ImageNet/Pretrained_model/train_only_knowns.txt',
# delimiter=' ',header=None)#,memory_map=True)#, nrows=10000)
if args.run_on_logits:
# results_file=np.array(results_file)
print results_file.shape,results_file[:,1:].shape
feature_vector_file=results_file[:,1:]
"""
for a in range(100):
t=cls_mav_dist(a)
"""
p = Pool(48)
MAV_models = p.map(cls_mav_dist,range(10))
p.close()
p.join()
data = dict(zip(range(100),MAV_models))
# cPickle.dump(data,open('/home/adhamija/faceness/ImageNet/Pretrained_model/Models_data','wb'))
#print MAV_models
del feature_vector_file
del results_file
#else:
# data = cPickle.load(open('/home/adhamija/faceness/ImageNet/Pretrained_model/Models_data','rb'))
def openmax_processing(row,ind=0):
feature_to_process = row[1:].tolist()
# print "feature_to_process",len(feature_to_process)
# print row.name,row[0],results_file.iloc[row.name,0]
# print row.name,results_file.shape
raw_decision = results_file.iloc[row.name,2:].tolist()
# raw_decision = results_file[results_file[0]==row[0]][2:]
# unknown_score=0
# print "raw_decision[0]",len(raw_decision),raw_decision[0]
unknown_score=max(raw_decision[0],0)
ranks = np.argsort(raw_decision)[::-1]
weights = np.zeros(len(raw_decision))
wscores=[]
for i, r in enumerate(ranks):
weights[r] = max((alpha-1.-i) / (alpha-1.), 0.)
MAV = data[i]['MAV'].tolist()
#print "MAV",MAV
model = data[i]['model']
#print "model",model
if args.use_euclidean:
distance = scipy.spatial.distance.euclidean(MAV, feature_to_process)
else:
distance = scipy.spatial.distance.cosine(MAV, feature_to_process)
wscores.append(model.w_score(distance))
weights = 1. - wscores * weights
revised_decision = raw_decision * weights
unknown_score += np.sum(raw_decision * (1.-weights))
pred=[unknown_score]+revised_decision.tolist()
softmax_scores=np.exp(pred)/np.sum(np.exp(pred))
return softmax_scores.tolist()
def process_each_split(ind):
return feature_vector_file.iloc[ind:ind+rows_per_proc].apply(openmax_processing,ind=ind,axis=1)
feature_vector_file = pd.read_csv('Test_MNIST_openMax_deep_features.txt',
delimiter=' ',header=None)#,memory_map=True)#, nrows=10000)
results_file = pd.read_csv('Test_MNIST_openMax_pred.txt',
delimiter=' ',header=None)#,memory_map=True)#, nrows=10000)
"""
feature_vector_file = pd.read_csv('/home/adhamija/faceness/ImageNet/Pretrained_model/features_val_knowns_and_unknowns.txt',
delimiter=' ',header=None)#,memory_map=True)#, nrows=10000)
results_file = pd.read_csv('/home/adhamija/faceness/ImageNet/Pretrained_model/val_knowns_and_unknowns.txt',
delimiter=' ',header=None)#,memory_map=True)#, nrows=10000)
"""
if args.run_on_logits:
feature_vector_file=results_file[:,1:]
print "feature_vector_file",feature_vector_file.shape
no_of_rows=feature_vector_file.shape[0]
rows_per_proc = no_of_rows/48
print range(0,no_of_rows,rows_per_proc)
p = Pool(48)
MAV_models = p.map(process_each_split,range(0,no_of_rows,rows_per_proc))
p.close()
p.join()
print "len(MAV_models)",len(MAV_models)
data_to_write=[]#pd.DataFrame()
for m in MAV_models:
print m.shape
data_to_write.extend(m.tolist())
data_to_write=np.array(data_to_write)
print data_to_write.shape
if args.use_euclidean:
# tools.write_file_for_DIR(results_file.iloc[:,1],data_to_write[:,1:],args.dir_file_name+'/euclidean_'+str(args.tail_size)+'_openmax.txt')
tools.write_file_for_DIR(
results_file.iloc[:,1],
data_to_write[:,1:],
('OpenMax/{}_euclidean_'+str(args.tail_size)+'_openmax.txt').format('DIR_with_UK'),#args.dir_file_name),
feature_vector=None,num_of_known_classes=10
)
else:
print results_file.iloc[:,1].shape,data_to_write[:,1:].shape
tools.write_file_for_DIR(
results_file.iloc[:,1],
data_to_write[:,1:],
# args.dir_file_name+'/cosine_'+str(args.tail_size)+'_openmax.txt'
('OpenMax/{}_cosine_'+str(args.tail_size)+'_openmax.txt').format('DIR_with_UK'),#(args.dir_file_name),
feature_vector=None,num_of_known_classes=10
)