Question

我非常混淆了使用SHAP解释模型时与分类栏做些什么。

As for example, I m using the bike sharing dataset from kaggle (https://www.kaggle.com/datasets/lakshmi25npathi/bike-sharing-dataset)

在对分类栏(即天气/周日)进行热编码之后,对模型进行了培训,并与SHAP做了解释,所解释的特点与一个热编码特征相同。

Is there a way or concept on how to merge back the SHAP value of those one hot encoded feature into single feature? (for example, SHAP value of one hot encoded feature: Weather_Spring, Weather_Summer, Weather_Fall, Weather_Snow back into single column value of Weather)

Answer 1

是:SHAP值是附加值。因此,你可以增加所有分类元变量的SHAP值,即X美元,以达到其水平的SHAP值。

在实践中,人们很少提到问题:

Tree based models usually offer other approaches to deal with categoricals (integer encoding, internal algorithms, internal OHE). There is no need for doing the OHE yourself.
For non-tree based models that require OHE, you can pack the preprocessor and the model into a Scikit-Learn pipeline. Then, you also don t have the problem as the SHAP explainer will "see" the original input, not the dummies.

友情链接