Distances derivation, explain_instance() method in Tabular LIME

[SamiRawash]

Hi,

I have a question regarding the explain_instance() method in the LimeTabularExplainer class. When using the LIME formula, we should compute the kernel using the distances between the points in the "real" space, not between their interpretable representation (where the points are in {0,1}^(number of features)), right?

The __data_inverse() method in the class returns the matrices data (interpretable representation) and inverse (real representation). In explain_instance(), however, we compute the distances using the rescaled data matrix, not the rescaled inverse matrix. Why is this the correct thing to do? What am I missing?

Thanks!