-
Notifications
You must be signed in to change notification settings - Fork 7
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Integration with MeasureTheory.jl #8
Comments
Very nice package! I've been looking for a proper way to deal with finite discrete measures for a while, since there is no such type of distribution in Distributions.jl. Now, am I mislead by the name or is this implementation not "well suitable" for actually multivariate distributions, I mean, when instead of unordered labels we have something like many samples from R^n? |
Hi @ablaom , thanks for the ping :) Generally, for some
It's valid for a measure to have itself as a base measure, which would make that measure "primitive". In this case we should get a log-density of 0.0. The interface is still in development, and I'd welcome collaboration. "Primitiveness" should probably be trait-based. If
In any case, in MeasureTheory we usually have a logpdf(cd::CategoricalDistribution) = datadependentterms(cd) + normalizationterms(cd) though probably with different names ;) Splitting things up in this way makes it easy to optimize product measures, pulling the normalization terms of of the loop. |
@cscherrer Thanks for that. @davibarreira No, you are not mislead. This is not yet multivariate. |
Thanks for the answer @ablaom. |
The atomic objects defined in this package are just non-negative measures over a labelled sample space (see here) and so ought to fit into MeasureTheory.jl framework.
@cscherrer Be great if you can give a run-down of what's required. This package is a port of functionality still in MLJBase but with plans to replace it. While it's now publicly available, I've not promoted it all and there's scope for fixing things you may not like.
The text was updated successfully, but these errors were encountered: