def TestOptF():
#if __name__ == "__main__":
'''
result:
LR:
[[ 915. 97.]
[ 267. 223.]]
0.834092980857 0.550617283951
0.692355132404
[[ 970. 94.]
[ 204. 233.]]
0.866845397676 0.609947643979
0.738396520828
[[ 932. 102.]
[ 259. 208.]]
0.837752808989 0.535392535393
0.686572672191
[[ 884. 62.]
[ 307. 248.]]
0.827328029949 0.573410404624
0.700369217286
[[ 919. 75.]
[ 292. 215.]]
0.833560090703 0.539523212045
0.686541651374
Avg F1: 0.700847038817
OptF:
[[ 721. 291.]
[ 130. 360.]]
0.774020397209 0.631025416301
0.702522906755
[[ 794. 270.]
[ 116. 321.]]
0.804457953394 0.624513618677
0.714485786036
[[ 709. 325.]
[ 130. 337.]]
0.757074212493 0.596988485385
0.677031348939
[[ 719. 227.]
[ 160. 395.]]
0.787945205479 0.671197960918
0.729571583199
[[ 719. 275.]
[ 158. 349.]]
0.768572955639 0.617152961981
0.69286295881
Avg F1: 0.703294916748
conclusion:
F1 on O is promoted, but F1 on N is decreased
after avg, no obvious diff
'''
with open('../../data2/features_all.pkl', 'rb') as my_input:
all_pid = dill.load(my_input)
all_feature = dill.load(my_input)
all_label = dill.load(my_input)
### preprocess
all_feature = np.array(all_feature)
selected = [i for i, x in enumerate(all_label) if x == 'N' or x == 'O']
all_label = np.array(all_label)
all_feature = all_feature[selected]
all_label = all_label[selected]
all_label_num = np.array(ReadData.LabelTo2(all_label, 'O'))
## k-fold cross validation
# F1_list = []
# kf = KFold(n_splits=5)
# for train_index, test_index in kf.split(all_label):
# train_data = all_feature[train_index]
# train_label = all_label[train_index]
# train_label_num = all_label_num[train_index]
# test_data = all_feature[test_index]
# test_label = all_label[test_index]
# test_label_num = all_label_num[test_index]
#
# clf = LRSimp()
# clf.fit(train_data, train_label)
# pred = []
# n_row, n_col = test_data.shape
# for i in range(n_row):
# pred.extend(clf.predict(list(test_data[i])))
# # break
# F1_list.append(MyEval.F1Score2(pred, test_label))
#
# print('\n\nAvg F1: ', np.mean(F1_list))
F1_list = []
kf = KFold(n_splits=5)
for train_index, test_index in kf.split(all_label):
train_data = all_feature[train_index]
train_label = all_label[train_index]
train_label_num = all_label_num[train_index]
test_data = all_feature[test_index]
test_label = all_label[test_index]
test_label_num = all_label_num[test_index]
test_data = sklearn.preprocessing.scale(test_data, axis=0)
clf = OptF()
clf.fit(train_data, train_label_num)
pred = []
n_row, n_col = test_data.shape
for i in range(n_row):
pred_prob = clf.predict_prob(list(test_data[i]))[0]
if pred_prob > 0.5:
pred.append('O')
else:
pred.append('N')
# break
F1_list.append(MyEval.F1Score2(pred, test_label))
print('\n\nAvg F1: ', np.mean(F1_list))
