Richard Chukwu

Results 14 comments of Richard Chukwu

add optimization subpackage in `functorch`

You're right alibool. Reuben Feinman's library Pytorch-minimize is awesome. Functorch should have something equivalent to jsoptimize

Distribution on pypi

I would like to ask if there is any update on the release via pypi. If not I would like to know how best to use torchmin as a dependency...

Distribution on pypi

Thanks @rfeinman, if I run the above command it successfully installs ``` #Collecting pytorch-minimize #Downloading https://test-#files.pythonhosted.org/packages/49/d3/ba8743747af15e8298a3c06bdb0cfad58ff09ddd8dc3cd8c9f4d5ff47a37/pytorch_minimi#ze-0.0.1-py3-none-any.whl (29 kB) #Installing collected packages: pytorch-minimize #Successfully installed pytorch-minimize-0.0.1 ``` Then upon running import...

Distribution on pypi

I've never uploaded to pypi so I'm at a loss here. Maybe @calvinmccarter can help.

Distribution on pypi

YaaaY. It works

Output the hess_inv from bfgs using the Optimizer API

I found it in optimizer._result Thanks

Different results between jaxopt.LBFGS and jaxopt.ScipyMinimize(method - 'l-bfgs-b')

I also did try with the zoom line search and did not get the correct results. You're right the problem is nonconvex.

Different results between jaxopt.LBFGS and jaxopt.ScipyMinimize(method - 'l-bfgs-b')

Nevertheless, I found that I could use list(map(func, *args)) instead of vmap with jaxopt.scipy.minimize and it temporarily solves my problem

Problem with vmapping rootfinder

Modifying the last line to use map instead of vmap works ``` import jax from jax.config import config; config.update("jax_enable_x64", True) import jax.numpy as np from jax import random,vmap from jaxopt...

Problem with vmapping rootfinder

You're right map is slower. That is why it is a tenmporary fix.