NicolasHug/Surprise

measures param for grid search and cross_validate should support callables

Open

#166 opened on Apr 6, 2018

View on GitHub
 (0 comments) (0 reactions) (0 assignees)Python (6,098 stars) (1,043 forks)batch import
enhancementgood first issue

Description

For now only strings are accepted as the measures parameter in GridSearchCV, RandomizedSearchCV, and cross_validate. It's thus impossible to use those with measures that take specific parameters as input (e.g. #156 ), or to use custom measures.

We should then accept callables in addition to strings.

Each callable should only take the predictions parameter. In order to handle measures with mulptile parameters, we could implement a make_measure helper (much like sklearn make_scorer) which would simply perform some partial application on the parameters, and set a greater_is_better parameter (this would clean a bit the current code).

Any other option is welcome

Contributor guide

Tech stack
python
Domain
datamachine learning
Issue type
feature
DifficultyEstimated implementation difficulty for a new contributor, from 1 for very small changes to 5 for expert-level work.
3
Estimated timeA rough time range for an experienced contributor to investigate, implement, test, and prepare a pull request.
1-2 days
Activity statusHow available the issue appears right now: fresh, active, stale, blocked, or waiting on maintainer input.
stale
ClarityHow clearly the issue explains the expected change, acceptance criteria, and next step.
mostly clear
Prerequisites
PythonFamiliarity with Surprise libraryUnderstanding of callables and partial application
Newbie friendlinessA 1-100 score estimating how approachable this issue is for first-time contributors.
60
Research direction
Investigate the current implementation of GridSearchCV, RandomizedSearchCV, and cross validate in the Surprise source code, likely in grid search.py and evaluate.py. Understand how the 'measures' parameter is validated and processed. Then implement support for callables by creating a make measure helper function similar to sklearn's make scorer, which accepts a callable with only 'predictions' parameter and sets a 'greater is better' attribute. Update the classes to accept both strings and callables, and consider extending the helper to handle partial application for measures with multiple parameters.