如下圖，SHAP支持模型解釋性，可用于抽取特征對結果的貢獻。

?Github地址：https://github.com/slundberg/shap

案例：

1、數(shù)據集：

# dataset：http://scikit-learn.org/stable/modules/generated/sklearn.datasets.load_diabetes.html # diabetes 是一個關于糖尿病的數(shù)據集， 該數(shù)據集包括442個病人的生理數(shù)據及一年以后的病情發(fā)展情況。 # 數(shù)據集中的特征值總共10項, 如下:# 年齡# 性別#體質指數(shù)#血壓#s1,s2,s3,s4,s4,s6 (六種血清的化驗數(shù)據) #10個t特征每個都做了均值中心化處理，然后又用標準差乘以個體數(shù)量調整了數(shù)值范圍。驗證就會發(fā)現(xiàn)任何一列的所有數(shù)值平方和為1.

2、代碼：

#https://github.com/slundberg/shap #https://slundberg.github.io/shap/notebooks/plots/decision_plot.html import lightgbm as lgb import shap# train XGBoost model diabetes=datasets.load_diabetes() x_train,x_test,y_train,y_test=train_test_split(diabetes.data,diabetes.target,test_size=0.1, random_state=10) print(type(x_test)) d_train = lgb.Dataset(x_train, label=y_train) d_test = lgb.Dataset(x_test, label=y_test)params = {"max_bin": 512,"learning_rate": 0.05,"boosting_type": "gbdt","objective": "regression","metric": "regression_l2","num_leaves": 10,"verbose": -1,"min_data": 100,"boost_from_average": True,"random_state": random_state }model = lgb.train(params, d_train, 10000, valid_sets=[d_test], early_stopping_rounds=50, verbose_eval=1000)# explain the model's predictions using SHAP explainer = shap.TreeExplainer(model) expected_value = explainer.expected_value if isinstance(expected_value, list):expected_value = expected_value[1] print(f"Explainer expected value: {expected_value}")with warnings.catch_warnings():warnings.simplefilter("ignore")shap_values = explainer.shap_values(x_test)[1]shap_interaction_values = explainer.shap_interaction_values(x_test) if isinstance(shap_interaction_values, list):shap_interaction_values = shap_interaction_values[1]features_display = ['Age','Gender','BMI','BloodPressure','S1','S2','S3','S4','S5','S6'] shap.decision_plot(expected_value, shap_values, features_display)

?更多圖示可參考，目前說支持xgboost、lgbm、sklearn以及深度keras\tf模型，深度學習模型還沒試驗，xgbosst在1.0.0版本據說支持，我安裝的是1.2.0不支持，sklearn我用lr也不支持。

總結

以上是生活随笔為你收集整理的机器学习模型解释性工具SHAP的全部內容，希望文章能夠幫你解決所遇到的問題。

如果覺得生活随笔網站內容還不錯，歡迎將生活随笔推薦給好友。