search

mcleonard / sampyl

MCMC samplers for Bayesian estimation in Python, including Metropolis-Hastings, NUTS, and Slice
http://mcleonard.github.io/sampyl/
MIT License
329 stars 53 forks source link

Is Sampyl still being developed/maintained? #18

Open Gabriel-p opened 6 years ago

Gabriel-p commented 6 years ago

Hi, I'm looking for a simple to use Bayesian MCMC sampler fully Python based. I've been trying emcee and now I've found Sampyl, which looks really good.

Before attempting to implement it though, I'd like to know if it is still being developed or at least maintained. For what I can see the latest commit is 1.5 years old with no active branches, and the opened issues are ~2 years old with no apparent progress in either of them.

Thanks.

twiecki commented 6 years ago

Have you looked at PyMC3?

ipashchenko commented 6 years ago

Here the author explains why he has implemented sampyl and what is a problem with pymc3: http://matatat.org/ab-testing-with-sampyl.html

Gabriel-p commented 6 years ago

@twiecki I have, but as @ipashchenko comments it depends on an annoying syntax to write your models. It's the same issue with PyStan, and that's why Sampyl looks so interesting.

twiecki commented 6 years ago

Yeah, theano makes things tricky but I'm always surprised that people would rather code their own likelihoods and model logps.

ipashchenko commented 6 years ago

@Gabriel-p , there's nice and easy to use python implementation of the nested sampling here by @kbarbary. So if you are not sticked to MCMC and want to get evidence estimates for e.g. model comparison - just check it.

Gabriel-p commented 6 years ago

@twiecki In my case my likelihood is already written, and it is not trivial. I'd have to invest the time into learning Theano (or PyStan's syntax) before even attempting to get a glimpse of how adequate the package is for my particular issue.

@ipashchenko I'll definitely look into it. Thank you.

madanh commented 6 years ago

@twiecki For me it's the opposite. I see the necessity to learn a domain specific language and a limited set of building blocks as the (annoying) price to pay for automatic differentiation and the speed-ups it brings. Perhaps this difference in views is because for people formally trained in statistic navigating the zoo of available distributions (and associated jargon) is easy while for people who learn Bayesian probability and MCMC to make inferences in their specific problems it is somewhat problematic.

Gabriel-p commented 6 years ago

Now even Theano is being discontinued too: https://groups.google.com/forum/#!msg/theano-users/7Poq8BZutbY/rNCIfvAEAwAJ

madanh commented 6 years ago

@Gabriel-p Did you try using sampyl as it is? I cloned it and and tried the examples from docs. Those wouldn't run due to a bug that I managed to fix/work around and now the examples are working fine. So if you want to try this now I can fork and push what I have. Will definitely play a bit more with this.

twiecki commented 6 years ago

Just submit a PR to this repo. Thanks!

On Jan 21, 2018 3:52 PM, "Hennadii Madan" notifications@github.com wrote:

@Gabriel-p https://github.com/gabriel-p Did you try using sampyl as it is? I cloned it and and tried the examples from docs. Those wouldn't run due to a bug that I managed to fix/work around and now the examples are working fine. So if you want to try this now I can fork and push what I have. Will definitely play a bit more with this.

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/mcleonard/sampyl/issues/18#issuecomment-359253972, or mute the thread https://github.com/notifications/unsubscribe-auth/AApJmPqQwXsK9gb611M_RlXRm6kgaCBgks5tM08egaJpZM4QQDOv .

madanh commented 6 years ago

@twiecki Thomas, just wondering: do you have permissions to merge, or you just want to try?

twiecki commented 6 years ago

I don't and actually confused this repo with "sampled" so disregard my earlier comment. Still think a PR couldn't hurt though :)

ipashchenko commented 6 years ago

@madanh , what is the problem? Check here Edit: Oops, it is your's PR...

madanh commented 6 years ago

@ipashchenko I could not quite understand your message. Care to elaborate? @twiecki Happens :) Anyway, here's the PR. https://github.com/mcleonard/sampyl/pull/19

Gabriel-p commented 6 years ago

@madanh I have not used sampyl yet, I found while scavenging for Python MCMCs and thought it looked interesting, which is why I am interested in knowing about its development.

madanh commented 6 years ago

@Gabriel-p Apparently is is not maintained, so it is up to us to pick up the banner.

mcleonard commented 6 years ago

@madanh @Gabriel-p Hi, I'm still sort of working on this and actually made a few updates a couple weeks ago.

This was a project I started in my postdoc year for my own work. Then I started working at Udacity and stopped doing much data analysis myself. So this kind of fell out of my free time work.

Seems like there is serious interest in using Sampyl, so I could be convinced to pick it back up. I do really love Bayesian data analysis.

madanh commented 6 years ago

Good to have you back and nice work BTW. Do you have something to apply sampyl to now. It's my impression is that it's hard to find motivation to maintain something that does not help solve ones own everyday problems (unless one gets paid for this).

zoj613 commented 4 years ago

any updates on the development of this project?

jeremysanders commented 3 years ago

Would it be possible to give someone else write access to at least fix the obvious bugs?