Custom metric

I'm having difficulties to implement the precision at k for a custom model. My model consists of two component a lightfm model + a custom model. The custom model takes the scores of lightfm and transforms them. As precision at k is computed directly from ranks of lightfm. I did not find a way compute them using lightfm model + custom model

Any help Thank you

Start by reviewing how LightFM computes precision at k from model ranks, then compare that flow with the custom model's transformed scores. Clarify whether the combined model should expose new scores or ranks, and define completion as precision at k being computable from the LightFM plus custom-model outputs.

Written by the indexing model from the issue text.