search

ContinuumIO / elm

Phase I & part of Phase II of NASA SBIR - Parallel Machine Learning on Satellite Data
http://ensemble-learning-models.readthedocs.io
44 stars 24 forks source link

Fixes for multiobjective optimization #222

Open PeterDSteinberg opened 6 years ago

PeterDSteinberg commented 6 years ago

For this issue, make sure that EaSearchCV (and other elm.model_selection classes) can score with cross validation when >1 scoring callable is given. Use dask-ml capabilities from PR 68

PeterDSteinberg commented 6 years ago

Hi @TomAugspurger - saw your merge here : https://github.com/dask/dask-searchcv/pull/60

It's closely related to this Elm issue. We need the evolutionary algo's pareto sorting to take advantage of the multiple scoring functions support from that PR. I assigned this one to you - it is lower priority than the demos with spectral clustering.

TomAugspurger commented 6 years ago

Thanks,

For getting up to speed, is there a place in the repo that currently does single-objective optimization?

PeterDSteinberg commented 6 years ago

elm.model_selection.ea_search - EaSearchCV wraps the dask-searchcv base class and its fit method elm.model_selection.evolve - general wrapper for EA algo based on deap's NSGA-2 example; has a generator that yields new parameter sets