train]
### SVM ###
# kernel:‘linear’, ‘poly’, ‘rbf’, ‘sigmoid’, ‘precomputed’
clf = svm.SVC(C=10, gamma=0.01, kernel='rbf', probability=True)
clf.fit(X_train_stats_kfold, y_train_kfold)
prediction = clf.predict(X_test_stats_kfold)
prob = clf.predict_proba(X_test_stats_kfold)
prob = prob[:, 1]
print(prob, prediction)
fpr, tpr, thresholds = roc_curve(y_test_kfold, prob)
print(type(fpr))
roc_auc = auc(fpr, tpr) #auc为Roc曲线下的面积
fpr = fpr.tolist()
tpr = tpr.tolist()
print('FPR, TPR: ', fpr, '\n', tpr)
print('auc:', roc_auc)
_precision, _recall, _f1_score, _acc, _TNR = getAccuracy(
prediction, y_test_kfold)
print('precision:', _precision, 'recall', _recall, 'f1_score',
_f1_score, 'accuracy', _acc, 'TNR', _TNR)
kfold_precision.append(_precision)
kfold_recall.append(_recall)
kfold_f1_score.append(_f1_score)
kfold_acc.append(_acc)
kfold_TNR.append(_TNR)
fold_counter += 1
print('10 k average evaluation is:',
'precision:', np.mean(kfold_precision), 'recall',
np.mean(kfold_recall), 'f1_score', np.mean(kfold_f1_score),
'accuracy', np.mean(kfold_acc), 'TNR', np.mean(kfold_TNR))
test]
X_val_stats_kfold, y_val_kfold = X_train_stats[
train[-1000:]], y_train[train[-1000:]]
X_train_stats_kfold, y_train_kfold = X_train_stats[
train[:-1000]], y_train[train[:-1000]]
# 采用后1000条做验证集
# X_val, y_val = X_train[-1000:], y_train[-1000:]
# X_train, y_train = X_train[:-1000], y_train[:-1000]
### Decision Tree ###
### xgboost ###
y_test, y_pred = xgboost_model(X_train_stats_kfold, y_train_kfold,
X_test_stats_kfold, y_test_kfold,
max_depth, min_child_weight)
print(y_test, y_pred)
_precision, _recall, _f1_score, _acc, _TNR = getAccuracy(
y_pred, y_test)
print('precision:', _precision, 'recall', _recall, 'f1_score',
_f1_score, 'accuracy', _acc, 'TNR', _TNR)
kfold_precision.append(_precision)
kfold_recall.append(_recall)
kfold_f1_score.append(_f1_score)
kfold_acc.append(_acc)
kfold_TNR.append(_TNR)
fold_counter += 1
time_end = time.time()
_time = time_end - time_start
kfold_time.append(_time)
print('10 k average evaluation is:', 'precision:',
np.mean(kfold_precision), 'recall', np.mean(kfold_recall),
