Aesara, theano, theano-pymc

Hi there,

I’m a little bit confused with aesara and whether I should update my programs that use PyMC3 (3.11) and additional theano capabilities or just wait.

Basically I’m wondering whether I can just replace theano.tensor and theano.shared (mostly basic functionalities) with the aesara equivalents in a seamless way, and thus be ready for the next PyMC3 updates.

Also, with aesara+JAX, I understand one can hope some performance improvement even with no GPU, is that correct?



I am on the same struggle. Should I update or remain with my old version??

It is recommended to update to pymc3.11.2, which relies on theano-pymc (not aesara). This is the latest stable version of pymc3.

The next version, using aesara will still take some time to be "production’ ready