Features: Automatically detect pipeline hyperparameter benchmark.

[alexandrebarachant]

trafficstars

The moabb can be used to benchmark a parameters search across multiple dataset. For example, one might be interested to varie the distance metric for a MDM algorithm, or the number of spatial filters. This can be done by providing to the moabb identical pipeline with different parameters.

One cool feature we can add is to detect that and create a column for the parameter in the results dataframe.

[vinay-jayaram]

I like it! It would be interesting to see if 6 really is the magic number for CSP filters, among other things. But this would make the current results structure really difficult, I think -- maybe it's worth adding with the database results PR? (...I will update that soon)

[vinay-jayaram]

I think this is technically solved, in that the feature is possible yes?

[sylvchev]

Could use the additional columns #127

[bruAristimunha]

Hi @alexandrebarachant, and @vinay-jayaram!

Can we close this PR?

[sylvchev]

Nice idea but it will raise the code complexity of evaluation a lot. A more maintainable solution is to dig the traces from GridSearch folder and make a dedicated function to retrieve the value from a given parameter, something like result_dataframe = get_grid_search_parameter(path="./results", params=["csp__n_filters", "svm__kernel"])