Project

General

Profile

Packaging Request #3461

Aesara

nona - about 1 year ago - . Updated about 1 year ago.

Status:
open
Priority:
wish
Assignee:
-
% Done:

0%

Description

Hello,

Aesara seems like a free alternative for multi-dimensional array manipulation and differentiation a.k.a. tensor manipulation (which can be used for deep learning). It is based on Theano and PyMC.

https://aesara.readthedocs.io/en/latest/introduction.html

Related issues

Related to Packages - Freedom Issue #3110: [python-pytorch] requires intel-mkl (unfree)fixed

Actions
Related to Packages - Freedom Issue #3326: [python-pytorch-opt-cuda] Depends on proprietary, blacklisted packagesfixed

Actions
Related to Packages - Packaging Request #2599: [tensorflow] depends on non-free bazelopen

Actions

History

#1

Updated by bill-auger about 1 year ago

  • Priority changed from bug to wish

an alternative to what though? - is there a blacklisted package that this could replace?

#2

Updated by nona about 1 year ago

I didn't want to mention PyTorch, TensorFlow, CNTK and all its derivatives. PyTorch is still in the repos, but some time ago it was added to the very long list of software which needs to be kicked out, a.k.a. blacklisted (or preemptively checked). In that particular case, it cannot be compiled without intel-mkl. Scikit-Learn is not an alternative for the type of gradient optimisation which is needed for CNN or deep learning (quote needed, but I read it on their own website--it has to do with GPU).

Translation: PyTorch is not yet black-listed. It is in the list to be checked. It can only be compiled with intel-mkl.

Aesara seems to be a successor of Theano (development stopped circa 2018). PyMC will fork Aesara into PyTensor and we may end with an alternative.

Cheers, bill. Nice to read you again.

#3

Updated by bill-auger about 1 year ago

#4

Updated by bill-auger about 1 year ago

  • Related to Freedom Issue #3326: [python-pytorch-opt-cuda] Depends on proprietary, blacklisted packages added
#5

Updated by bill-auger about 1 year ago

#6

Updated by nona about 1 year ago

For completeness, Aesara does not support GPU at the moment:

device (cpu)
      Doc:  Default device for computations. only cpu is supported for now
      Value:  cpu

Also: https://discourse.pymc.io/t/gpu-acceleration-with-aesara/9814, https://github.com/pymc-devs/pymc/issues/5894

Also available in: Atom PDF