Another approach is to use variational inference rather than sampling. If you call pm.fit instead of pm.sample that will run VI, which will be fast (though not as good of an estimate as you’d get from MCMC, in general).
Another approach is to use variational inference rather than sampling. If you call pm.fit instead of pm.sample that will run VI, which will be fast (though not as good of an estimate as you’d get from MCMC, in general).