Describe the workflow you want to enable

I would like to save off the results of all runs in GridSearchCV to MLFlow. MLFlow

for param in params:
    with mlflow.start_run():
        est = ElasticNet(**param)
        est.fit(train_x, train_y)
        metrics = est.score(test_x, test_y)
        mlflow.log_params(param)
        mlflow.log_metrics(metrics)
        mlflow.sklearn.log_model(est, "model")

See https://mlflow.org/docs/latest/tutorials-and-examples/tutorial.html for more details:

I would like to use GridSearchCV to do the above because it comes with many other features (e.g. HalvingGridSearchCV, multi-threading, etc ...)

Describe your proposed solution

A callback parameter to GridSearchCV. Perhaps

def log_candidate(model, test_x, test_y):
  with mlflow.start_run():
        mlflow.log_params(model.get_params())
        mlflow.log_metrics(metrics)
        mlflow.sklearn.log_model(est, "model")

Describe alternatives you've considered, if relevant

To hack the scorer for this purpose: https://danielhnyk.cz/adding-callback-to-a-sklearn-gridsearch/

This is suboptimal because:

1. If you want to return multiple metrics, you cannot save multiple scores using the provided API. This is because we have to pass multiple scorers, not a function that generates multiple scores.
2. Enabling return_train_score will call the scorer callback too many times and it is not easy to distinguish between the training and testing scoring.

Additional context

No response