Theano-JAX

196 views
Skip to first unread message

Thomas Wiecki

unread,
Aug 24, 2020, 12:25:51 PM8/24/20
to theano...@googlegroups.com
Hi all,

We, the PyMC development team have forked Theano and will continue to develop it: https://github.com/pymc-devs/Theano-PyMC

We have already done a lot of work in terms of changing the testing framework to pytest, fixing stylistic issues, and applying black formatting.

The most excited new development is adding a new JAX linker next to the cython/c one: https://github.com/pymc-devs/Theano-PyMC/pull/21

This turns out to be quite straightforward. If you would like to help with any aspect of this, we'd greatly appreciate it. Or if you have any PRs that never got merged into Theano, please consider resubmitting them there.

I think Theano is still an amazing framework that's mature, performant, and has a very readable and hackable code-base.

Best,
Thomas

Wong Hang

unread,
Aug 25, 2020, 2:42:07 AM8/25/20
to theano...@googlegroups.com
Hi there,

Will your team maintain libgpuarray as well? I have a bug fix in pygpu but haven't been merged:


Due to my employment arrangement, I need to seek approval before I can submit any new PR to open source projects. Would you mind to merge and fix it?
I don't care about the credits.

Best regards,
wonghang

Thomas Wiecki <thomas...@gmail.com> 於 2020年8月25日 週二 上午1:25寫道:
--

---
You received this message because you are subscribed to the Google Groups "theano-users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to theano-users...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/theano-users/CAPcXcF1QEnHrMn6d%3DQQ5Erkw--uCSeTxOE8WGSzG9ueuVCHz-g%40mail.gmail.com.

Paul Baggenstoss

unread,
Aug 25, 2020, 5:13:46 AM8/25/20
to theano-users
Hello,

It is nice to hear that Theano will be further improved.

By the way,  I have created highly GPU-paralleled routines having to do with solving symmetric positive-definite matrix equations,
and auto-regressive systems (Levinson). It also has gradients. .They are coded in C and compiled with CUDA, then linked into Theano using
the Magma interface. (basically, I coded in C, compiled using CUDA, linked them to libmagma.so, added the headers to magma.h
and some interface code to skcuda/magma.py. I also added Python class definitions in gpuarry/linalg.py and corresponding C interface
code in gpuarray/c_code ). I think it is very useful and fast code and fixes some deficiencies in Theano:  In Theano there is no fast and
fully parallel GPU implementation for solving and/or getting the determinant of symmetric positive-definite systems. 
An example would be to compute the probability distribution of a multi-variate Gaussian distribution - given its symmetric
positive-definite covariance matrix - on a batch of data - very slow in Theano using scan, and not GPU parallel.  

Perhaps there is some way to merge them intoTheano, but I'd need help in that. I would suggest
adding them as a separate library, similar to magma. I could provide the code and the interface code.

Paul Baggenstoss
Reply all
Reply to author
Forward
0 new messages