python - 如何计算可靠性图的置信度?
问题描述
我想校准我的分类模型,解释置信度直方图和可靠性图。我尝试在 Github 上使用Hollance 的方法,但如何计算可靠性图的置信度?
我尝试了以下代码(我使用了交叉验证):
def Get_Confidence(data, file):
X = data.drop(columns=['Label'])
y = data['Label']
DT = DecisionTreeClassifier(random_state=0)
RF = RandomForestClassifier(random_state=0)
XGB = XGBClassifier(random_state=0, verbosity = 0)
models = [('DT', DT), ('RF', RF), ('XGB', XGB)]
result_title = ['y', 'prediction', 'confidence']
if not os.path.isfile(file):
results = pd.DataFrame(columns = result_title)
else:
results = pd.read_csv(file)
for name, model in tqdm(models):
y_pred = model_selection.cross_val_predict(model, X, y, cv=5, method='predict')
y_proba = model_selection.cross_val_predict(model, X, y, cv=5, method='predict_proba')
row = pd.DataFrame([[y, y_pred, y_proba]], columns = result_title)
results = row
results.to_csv(file, index=False)
return
结果不是 Dataframe 显示那个图。
另外,我应该保存文件,因为我有各种数据。