Evaluation of recommendation files

Sometimes, the practitioner could need to evaluate an already computed recommendation file. Either we forgot to involve some metrics or we want to compare our models with external baselines, Elliot provides a facility to restore recommendation files and use them for the overall evaluation of the running experiment.

This is a sample config file with a proxy model restoring a recommendation file:

experiment:
    ...
    models:
        ProxyRecommender:
            path: path/to/recs/of/specific/model.tsv
        ItemKNN:
            ...

ProxyRecommender is a fake recommender model which is able to restore old recommendation and prepare all inner data structures to support Elliot evaluation pipeline.

Additionally, Elliot provides the practitioners with a facility to evaluate all the recommendation files stored in a folder.

This is a sample config file to restore recommendation files from a folder:

experiment:
    ...
    models:
        RecommendationFolder:
            folder: path/to/recs/folder
        ItemKNN:
            ...

RecommendationFolder is a fake recommendation model that restores all the recommendation files found in the target folder and prepare all inner data structures to support Elliot evaluation pipeline.