I am hearing a lot about probabilistic programming.

Turns out its just a way to specify probabilistic graphical models.

Like Tensorflow is to neural networks.

So, why use it? Do you know of any place where it has been useful?

  • $\begingroup$ Heard about stan? Or bugs? Or jags? $\endgroup$ – kjetil b halvorsen Mar 19 '18 at 7:27
  • $\begingroup$ Actually, tensorflow also aims to have the probabilistic interface, cf twitter.com/junpenglao/status/974379738365157378 $\endgroup$ – Tim Mar 19 '18 at 7:30
  • $\begingroup$ @kjetilbhalvorsen yes, they are frameworks that do the same thing -- kind of like DSLs for PPL right? But any success stories? Where are they used? Do they really make implemeniting PGMs easy? Any use other than PGMs? Any resources you could refer to? Thanks. $\endgroup$ – david nadal Mar 19 '18 at 7:31
  • $\begingroup$ @Tim great, not to learn too many thing $\endgroup$ – david nadal Mar 19 '18 at 7:32

If you ask about software, among the most popular probabilistic programming frameworks are BUGS, Jags, Stan, PyMC, INLA, and Edward.

If you ask about success stories, it has been used for many years by mediacal researchers, as described by Spiegelhalter et al in Bayesian Approaches to Clinical Trials and Health‐Care Evaluation, Andrew Gelman advocates their usage in social sciences, and Eric-Jan Wagenmakers and John Kruschke write much about their usage in psychology, physicists at CERN seem to use it, it is used by Google including the upcoming probabilistic interface in tensorflow, Facebook has released their time-series forecasting package that uses Stan, Uber has released their own probabilistic programming language and many, many others.

TLDR; it is very popular in many areas of research and industries.

See also the What is probabilistic programming? thread.

  • $\begingroup$ thanks, could you point to some algorithms that I run on these frameworks? What does this approach offer..I watched chris bishop's talk on model based machine learning which is just PGMs implemented with black box inference...PGMs are already sidelined right, they do not have much industrial relevance.. $\endgroup$ – david nadal Mar 19 '18 at 7:50
  • 1
    $\begingroup$ @davidnadal they do not run any particular algorithms (same as tensorflow does not), they are languages used for programming the probabilistic models that support MCMC sampling/or optimization using the models (same as tensorflow has build in SGD-like optimizers out-of-the-box). $\endgroup$ – Tim Mar 19 '18 at 7:53

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.