search

predict-idlab / powershap

A power-full Shapley feature selection method.
Other
193 stars 18 forks source link

Failed to reach convergence #24

Closed najeeb9522 closed 1 year ago

najeeb9522 commented 1 year ago

While running selector.fit(), I get the following warning. Failed to converge on a solution.

What does this indicate and what can be done to remedy this situation?

JarneVerhaeghe commented 1 year ago

This warning comes from the underlying SciPy library used to do the inverse power calculation to find the required number of iterations. This approach uses a heuristic algorithm and sometimes fails to converge if no number of required iterations can be found to reach the specified power requirement. If you are specifying the p-value and the power by yourself you can lower the power requirement to avoid the warning, however, if you are using the default values you can fully ignore this warning. If you are concerned about the effect of this warning, you can always look at the _processed_shaps_df in the selector by using selector._processed_shaps_df and inspect the features individually. Hopefully this answers your question!