This seems like a pretty easy win given how parallelizable the different chains are.
The current solution is to follow sklearn and pymc in using joblib to handle the parallelization. A nice feature of this is that it looks relatively easy to make this solution scale into a full dask cluster approach.
This seems like a pretty easy win given how parallelizable the different chains are.
The current solution is to follow sklearn and pymc in using joblib to handle the parallelization. A nice feature of this is that it looks relatively easy to make this solution scale into a full dask cluster approach.