jessegmeyerlab/positional-SHAP

for unsupervised model

Opened this issue · 1 comments

patience111 commented

Dear developers,

According to your description "To use PoSHAP for your own model, use the jupyter notebook present in the tutorial folder and follow the instructions inside. You will need a trained model, your x testing set, your x training set, your y training set, and a dictionary linking your encoded x values to the corresponding understandable input."
I am wondering is there any change the "positonal-SHAP" could be used to unsuperivised deep learning models such as autoencoder...

Thank you very much! 🙏

jessegmeyerlab commented

Since the concept is based on SHAP, which can interpret any model for which you have labels using kernelexplainer, then yes you should be able to do this. Your data labels will be the same as your data. Does that make sense?

  • 👍1