search

xwieme / MastatThesis

This repository contains all resources used for my master thesis in statistical data analysis.
0 stars 0 forks source link

Is SME model agnostic? #3

Closed xwieme closed 5 months ago

xwieme commented 1 year ago

SME is explained using one GNN architecture (relational GNN), does it perform the same if other GNN architectures are used? Since masking a substructure of a molecule does not require properties of the GNN model, it could also be used on other ML algorithms.

xwieme commented 8 months ago

SME and Shapley/HN values do not have any model assumptions, except for the possibility to partition a molecule in substructures. The explanations of different ML models with similar performance could be explored to see if the explanations are stable.