# clf = svm.SVC(C=0.1, kernel='rbf', gamma='auto',
# shrinking=True, probability=True, tol=0.0001,
# cache_size=1000, max_iter=-1, class_weight='balanced',
# decision_function_shape='ovr', random_state=None
# )
# clf=AdaBoostClassifier( n_estimators=150,algorithm='SAMME.R',learning_rate=0.8)#8,9
# clf=BaggingClassifier(n_estimators=200,max_samples=1.0,max_features=1.0,
# bootstrap=True,bootstrap_features=False,random_state=200)
clf.fit(train_in, train_out)
test_predict = clf.predict(test_in)
proba_test = clf.predict_proba(test_in)
train_predict = clf.predict(train_in)
proba_train = clf.predict_proba(train_in)
test1, test2 = ann.evaluatemodel(train_out, train_predict, proba_train[:,
1])
evaluate_train.extend(test1)
prenum_train.extend(test2)
test3, test4 = ann.evaluatemodel(test_out, test_predict,
proba_test[:,
1]) # test model with testset
evaluate_test.extend(test3)
prenum_test.extend(test4)
Result_test = pd.DataFrame(
evaluate_test, columns=['TPR', 'SPC', 'PPV', 'NPV', 'ACC', 'AUC', 'BER'])
# Result_test.to_csv('BER_LR_ks.csv')
Result_test.boxplot()
plt.show()
test_predict = sum_pre#测试集的预测结果
sum_prob = []
for j in range(np.shape(test_in)[0]):
colj = predict_prob[:, j]
if sum_pre[j] == 1:
tmp_prob = np.mean(colj[colj >= 0.5])
else:
tmp_prob = np.mean(colj[colj < 0.5])
# if tmp_prob == np.nan:
# print('test')
sum_prob.append(tmp_prob)
sum_prob = np.array(sum_prob)
proba_test = sum_prob#测试集概率预测结果
test3, test4 = ann.evaluatemodel(test_out, test_predict, proba_test) # test model with testset
evaluate_test.extend(test3)
prenum_test.extend(test4)
Result_test = pd.DataFrame(evaluate_test, columns=['TPR', 'SPC', 'PPV', 'NPV', 'ACC', 'AUC', 'BER'])
Result_test.to_csv('BER_LR_ks.csv')
Result_test.boxplot()
plt.show()
mean_test = np.mean(evaluate_test, axis=0)
std_test = np.std(evaluate_test, axis=0)
evaluate_test.append(mean_test)
evaluate_test.append(std_test)
evaluate_test = np.array(evaluate_test)
for train, test in skf.split(dataMat, labelMat):
print("%s %s" % (train, test))
train_in = dataMat[train]
test_in = dataMat[test]
train_out = labelMat[train]
test_out = labelMat[test]
train_in=train_in.reshape(-1,1)#只有一个特征值，过采样前特殊处理
train_in, train_out = RandomOverSampler().fit_sample(train_in, train_out)
thre_tmp=findthresh(train_in,train_out)
thre.append(thre_tmp)
train_predict = binaryclassify(train_in,thre_tmp)
test_predict=binaryclassify(test_in,thre_tmp)
test1, test2 = ann.evaluatemodel(train_out, train_predict, train_in) # test model with trainset
evaluate_train.extend(test1)
test3, test4 = ann.evaluatemodel(test_out, test_predict, test_in) # test model with testset
evaluate_test.extend(test3)
Result_test = pd.DataFrame(evaluate_test, columns=['TPR', 'SPC', 'PPV', 'NPV', 'ACC', 'AUC', 'BER'])
Result_test.to_csv('BER/BER_MEWS.csv')
Result_test.boxplot()
plt.show()
mean_train = np.mean(evaluate_train, axis=0)
std_train = np.std(evaluate_train, axis=0)
evaluate_train.append(mean_train)
evaluate_train.append(std_train)
skf = StratifiedKFold(n_splits=10) #十折交叉验证
kfold = 1
for train, test in skf.split(dataMat, labelMat):
print("第%s 次交叉验证：" % kfold)
train_in = dataMat[train]
test_in = dataMat[test]
train_out = labelMat[train]
test_out = labelMat[test]
train_in, train_out = RandomOverSampler().fit_sample(train_in,
train_out) #平衡测试集样本
train_predict, test_predict, proba_train, proba_test = ann.ANNClassifier(
neuo, train_in, train_out, test_in)
proba_train = proba_train[:, 1]
proba_test = proba_test[:, 1]
test1, test2 = ann.evaluatemodel(train_out, train_predict,
proba_train) #求训练集测试集的评价指标
evaluate_train.extend(test1) #训练集的各评价指标，百分比
prenum_train.extend(test2) #测试结果中，混淆矩阵对应的四个数
test3, test4 = ann.evaluatemodel(test_out, test_predict,
proba_test) #test model with testset
evaluate_test.extend(test3)
prenum_test.extend(test4)
kfold = kfold + 1 #十折交叉验证的次数
#对于测试集，保存模型预测的结果的评价指标，并绘制箱线图
Result_test = pd.DataFrame(
evaluate_test, columns=['TPR', 'SPC', 'PPV', 'NPV', 'ACC', 'AUC', 'BER'])
Result_test.to_csv('BER/BER_ANN_ks.csv')
Result_test.boxplot()
