I'm quite familiar with deep learning because it's so highly hyped up, but am less familiar with Probabilistic programming and Bayesian methods. So, I have a general question: is anyone using Probabilistic Programming in industry? Have people ditched it for DNNs? Are people taking hybrid approaches to try and mix the two?
The first time I saw this phrase I thought it was going to describe something like software branch prediction, speculative execution, self tuning algorithms, or heck even Bloom filters or hyperloglog. That was a direction my first mentor and I used to talk about and it’s one of my regrets that I never did much in that arena.<p>My brain wants this term to mean something else and I become momentarily excited every time this topic gets reposted.
A corresponding framework in Julia - Turing[1][2].<p><a href="https://turing.ml" rel="nofollow">https://turing.ml</a><p><a href="https://github.com/TuringLang" rel="nofollow">https://github.com/TuringLang</a>
It has been a few years since I looked at this, and it looks like a lot has been added since then. It's certainly worth a look. But at the time I found Allen B. Downey's "Think Bayes" to be a more thorough and comprehensive resource: <a href="https://greenteapress.com/wp/think-bayes/" rel="nofollow">https://greenteapress.com/wp/think-bayes/</a>
For anyone interested in learning more, Stan is an excellent alternative probabilistic programming language:<p><a href="https://mc-stan.org" rel="nofollow">https://mc-stan.org</a><p>with thorough documentation:<p><a href="https://mc-stan.org/users/documentation/" rel="nofollow">https://mc-stan.org/users/documentation/</a>
Oh, I've been meaning to go through this book as we used PyMC (I only did reviews and this went over my head a lot) on my last job to build an AB testing system.<p>I recently started going through it again and it's pretty fascinating as someone not familiar with the field.