ppscore when model_score>baseline_score

[bimhud]

Not sure why ppscore is still zero when model_score>baseline_score

[FlorianWetschoreck]

Thank you for reporting this. That really seems odd. Can you create a reproducible code example?

[bimhud]

@FlorianWetschoreck it's a standard call of the pps.predictors(df, "y"). The given dataframe contains small percentage of missing values. Numeric columns and category columns are also provided from the dataframe. No thing is really strange about the dataset. Unfortunately, it is unable to provide the dataset for re-producible. Just think about the what factor contributing to ppscore to make it become zero in this situation.

[FlorianWetschoreck]

Thank you for your reply.

I was confused before - the behavior is exactly what we want to have because:
For the mean absolute error, better means lower e.g. closer to 0. As you can see above, the baseline is lower than the model and thus, the PPS is 0 because the model is worse than the baseline.
This might be a little bit confusing because for the PPS higher is better but for MAE lower is better.

I hope that helps?