Parallel execution for h2o AutoML?

1
votes

Is it possible to kick off h2o's AutoML to train and evaluate models in parallel either locally or on a cluster, using something like joblib or dask?

1
h2o
You may have trouble finding someone that both known H2O and Dask well enough to answer this. Perhaps you should try something and report back specific problems you encounter. – mdurant

1 Answers

0
votes

Closing out this question as the question has migrated, and more appropriately, belongs elsewhere (i.e. the Google Group). Here's where the topic can be found. Also a reference to H2O posting guidelines here.

I'll recap the highlights here:

  • There's nothing outright that is known right now to do this, BUT:
  • H2O builds one model at a time, but the creation of that model is a parallel operation
  • You can potentially parallelize by having multiple API clients pointing to the same cluster, operating on different entries of in the search space