-
Notifications
You must be signed in to change notification settings - Fork 1
/
multilabel-labelset.py
263 lines (223 loc) · 8.23 KB
/
multilabel-labelset.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
#!/usr/bin/python2.7
import sys
from sklearn.model_selection import StratifiedKFold
import numpy as np
import arff
from skmultilearn.model_selection.measures import folds_label_combination_pairs_without_evidence
from skmultilearn.model_selection.measures import example_distribution
from skmultilearn.model_selection.measures import label_combination_distribution
from skmultilearn.model_selection.measures import folds_without_evidence_for_at_least_one_label_combination
class Transfomer:
def transform_to_multiclass(self, y):
self.label_count = y.shape[1]
self.unique_combinations = {}
self.reverse_combinations=[]
self.last_id = 0
train_vector = []
for labels_applied in y:
label_string = ",".join(map(str, labels_applied))
if label_string not in self.unique_combinations:
self.unique_combinations[label_string] = self.last_id
self.reverse_combinations.append(labels_applied)
self.last_id += 1
train_vector.append(self.unique_combinations[label_string])
return train_vector
def stratified_folds(n_splits, y):
t=Transfomer()
kf = StratifiedKFold(n_splits=n_splits, random_state=None, shuffle=False)
folds = [x[1] for x in list(kf.split(np.zeros(y.shape[0]),t.transform_to_multiclass(y)))]
return folds
# Call
if len(sys.argv) <= 2:
print "Correct use: multilabel-labelset.py input-file f [output-file-prefix]"
sys.exit()
f = int(sys.argv[2])
# Read arff file
if sys.argv[1].lower().endswith('.arff') == False :
sys.exit("Dataset format unknown, please use .arff datasets")
dataset = arff.load(open(sys.argv[1], 'rb'))
data = np.array(dataset['data'])
#We have to get the number of clases from the raw file
file = open(sys.argv[1], "r")
line = file.readline()
flag = False
for i in line.split():
if flag is True:
number = i
break
if (i == "-C") or (i == "-c"):
flag = True
if (flag==False):
file.close()
sys.exit("Wrong format for the dataset header")
if number[-1:] == "'":
number = number[:-1]
file.close()
#Now we have the number stored, knowing that positive means the first attributes and negative the last ones
nominalIndexArray = []
nominals = []
aux = 0
#from attributes we can get if its nominal
if int(number) > 0:
for x in dataset['attributes'][int(number):]:
if (len(x[1]) > 2) and (x[1] != ("NUMERIC" or "REAL" or "INTEGER" or "STRING")):
nominalIndexArray.append(aux)
nominals.append(x[1])
aux +=1
else:
for x in dataset['attributes'][:int(number)]:
if (len(x[1]) > 2) and (x[1] != ("NUMERIC" or "REAL" or "INTEGER" or "STRING")):
nominalIndexArray.append(aux)
nominals.append(x[1])
aux +=1
#Split the data in X and Y
if(int(number)>0):
y = data[:,0:int(number)].astype(int)
x = data[:,int(number):]
else:
y = data[:,int(number):].astype(int)
x = data[:,:int(number)]
if len(nominalIndexArray) > 0:
#Change the nominal attributes to numeric ones
index = 0
X = []
for k in x:
numericVector = []
for i in range(0, len(nominalIndexArray)):
#Ahora tenemos que crear el vector que le vamos a poner al final de cada
checkIfMissing = False
for aux in nominals[i]:
if aux == k[nominalIndexArray[i]]:
#Add 1 to the array
checkIfMissing = True
numericVector.append(1)
else:
#Add 0 to the array
checkIfMissing = True
numericVector.append(0)
if checkIfMissing is False:
#Add another 1 to the array
numericVector.append(1)
else:
numericVector.append(0)
auxVector = np.append(k, [numericVector])
#Substract that nominals values
auxVector = np.delete(auxVector, nominalIndexArray)
X.append(auxVector)
X = np.array(X)
else:
X = np.array(x)
# Sparse or dense?
sizeofdouble = 8
sizeofint = 4
sizeofptr = 8
dense_size = len(X)*len(X[0])*sizeofdouble+len(X)*sizeofptr
#nz = np.count_nonzero(X)
#Count_nonzero is not working so i'll count the non zeroes by myself (This will take a little longer but better than not counting them)
nz = 0
for i in range(0, len(X)):
for j in range(0, len(X[0])):
if X[i][j] != '0.0':
nz += 1
sparse_size = nz*(sizeofdouble+sizeofint)+2*len(X)*sizeofptr+len(X)*sizeofint
sparse = False if sparse_size >= dense_size else True
# Use input file as output suffix if no other given
suffix = sys.argv[3] if len(sys.argv) == 4 else sys.argv[1][:sys.argv[1].rfind('.')]
kfold = 0
folds = []
desired_number = []
# Labelset k-fold partition
kf = stratified_folds(f, y)
print ("Generating kfolds...")
for test_index in kf:
train_index = [x for x in range(X.shape[0]) if x not in test_index]
X_train, X_test = X[train_index], X[test_index]
y_train, y_test = y[train_index], y[test_index]
#Classes do not match
folds.append(train_index)
desired_number.append((X.shape[0]*(f-1))/f)
#Training file
fp = open(suffix+str(kfold)+'.ltrain', 'w')
#Save header
if sparse:
fp.write('[MULTILABEL, SPARSE]\n')
else:
fp.write('[MULTILABEL, DENSE]\n')
fp.write('$ %d\n' % len(X_train)) #Number of objects
fp.write('$ %d\n' % len(X_train[0])) #Number of attributes
fp.write('$ %d\n' % abs(int(number))) #Number of labels
#Data
for i in range(0, len(X_train)):
if sparse:
for j in range(0, len(X_train[i])):
if(X_train[i][j] != '0.0'):
fp.write(str(j+1)+':'+str(X_train[i][j])+' ')
if(X_train[i][j] == 'YES'):
fp.write('1'+' ')
else:
for j in range(0, len(X_train[i])):
if(X_train[i][j] == 'YES'):
fp.write('1'+' ')
elif (X_train[i][j] == 'NO'):
fp.write('0'+' ')
else:
fp.write(str(X_train[i][j])+' ')
fp.write('[ ')
for j in range(0, len(y_train[i])):
if y_train[i][j] == '0.0':
aux = str(y_train[i][j]).split('.')[0]
fp.write(str(int(aux))+' ')
else:
fp.write(str(int(y_train[i][j]))+' ')
fp.write(']\n')
fp.close()
#Testing file
fp = open(suffix+str(kfold)+'.ltest', 'w')
#Save header
if sparse:
fp.write('[MULTILABEL, SPARSE]\n')
else:
fp.write('[MULTILABEL, DENSE]\n')
fp.write('$ %d\n' % len(X_test))
fp.write('$ %d\n' % len(X_test[0]))
fp.write('$ %d\n' % abs(int(number)))
#Data
for i in range(0, len(X_test)):
if sparse:
for j in range(0, len(X_test[i])):
if(X_test[i][j] != '0.0'):
fp.write(str(j+1)+':'+str(X_test[i][j])+' ')
if(X_test[i][j] == 'YES'):
fp.write('1'+' ')
else:
for j in range(0, len(X_test[i])):
if(X_test[i][j] == 'YES'):
fp.write('1'+' ')
elif (X_test[i][j] == 'NO'):
fp.write('0'+' ')
else:
fp.write(str(X_test[i][j])+' ')
fp.write('[ ')
for j in range(0, len(y_test[i])):
if y_test[i][j] == '0.0':
aux = str(y_test[i][j]).split('.')[0]
fp.write(str(int(aux))+' ')
else:
fp.write(str(int(y_test[i][j]))+' ')
fp.write(']\n')
fp.close()
kfold += 1
fp = open(suffix+'.lmeasures', 'w')
FLZ = folds_label_combination_pairs_without_evidence(y, folds, 1)
FZ = folds_without_evidence_for_at_least_one_label_combination(y, folds, 1)
LD = label_combination_distribution(y, folds, 1)
ED = example_distribution(folds, desired_number)
fp.write("Label distribution: ")
fp.write(str(LD)+'\n')
fp.write("Example distribution: ")
fp.write(str(ED)+'\n')
fp.write("Number of fold-label pairs with 0 positive examples, FLZ: ")
fp.write(str(FLZ)+'\n')
fp.write("Number of folds that contain at least 1 label with 0 positive examples, FZ: ")
fp.write(str(FZ)+'\n')
fp.close()