Yea the shape is pretty cool. Hm, I don't think the author used the three models in succession. I think he just tried out different models to measure the result

Yea the shape is pretty cool. Hm, I don't think the author used the three models in succession. I think he just tried out different models to measure the results. Yea feature importance would've been cool too. I think the SHAP value does list the order of importance on the y-axis. Wonder if there's a way to extract more signal somehow?