What's your strategy regarding dependency on Theano?


#1

Hi there, since Theano won’t be developed anymore, I’m wondering whether to spend time learning this package to use PyMC3…

What’s your plan in the long run?

3ks


#2

Hi @KaneFu, the ongoing discussion is in https://discourse.pymc.io/c/pymc4, but depending on what is your use case, you might not need to learn a lot of Theano.


#3

Thanks for reply. I think I’ll try to implement my model first and see, just think it’s not worthwhile to learn theano from scratch, but Pytorch and Tensorflow seem to be worth the effort. 3ks!


#4

Yeah this might hurt PyMC in the long run by hemorrhaging users, specifically new ones, who want to avoid learning Theano. This is especially true now that there are well known libraries for both Tensorflow and Pytorch. Please don’t take my comment to imply impatience or anything like that. And believe me, I fully appreciate the fact that I am in no position to ask for anything from devs in an open source project who are working on a voluntary basis. Rather, as a well-wisher of the project, my comment comes from a good place.