Open tjad opened 4 years ago
@mrkn I can confirm that all existing functionality still works on 1.7.0. I have run some training using 1.7.0
This in preparation for 1.8.x update. I guess all that really needs to be done is extending the gluon api modules
I have further bumped versions in the ci. Perhaps you could publish a new docker image to facilitate those versions
@tjad Do you want to manage this project? If you can do it, I want to migrate this to red-data-tools organization and give you admin permission.
@mrkn Sounds like a plan. I'm eager to push the Ruby ML community forward with regards to tools.
As an aside, do you have considerations for proposing incubator-mxnet adopt this project, to officially support Ruby? I think they'll be pleased to have a 9th language aboard, if we can get the project quite updated (especially if it supports Gluon well - seeing as they want Gluon to be the framework of choice for "all languages" )
i worked on parts of the gluon support. i'd be happy to assist as well.
@tjad Sounds great. I'm OK to donate this repository to incubator-mxnet. Are the things below only remaining work to be done before the donation?
@toddsundsted Awesome!
@mrkn Yeah, I suppose they'd still rely on Red Data team to keep this repository updated anyway. As far as I can tell, yes completing above list would put this repository in great stead for MXNet adoption.
Still working towards proper solution for #48
@toddsundsted I think I mostly understood the transfer from python to ruby. It's simple and great work! And I hope we can push it forward for greater use.
I've added Adam optimiser in a optimiser
branch. Perhaps we could further branch out into the relevant modules of Gluon such as nn
, rnn
, loss
etc, then we can tackle outstanding functionality and merge back into master as completed.
I've opened #49 and #50 to help with coordination of tasks regarding coverage. When you're tackling a task, claim it by mentioning the item in the respective issue. As this is mostly a code transfer, we can emphasise on the implementations first, before test coverage is created. Other issues may arise out of implementing this coverage, such as stype
support
I think we can put the KV store coverage into the above list too WIP https://github.com/mrkn/mxnet.rb/commit/6c8df1b8cb0c63f9e711117af7823d7d1e2ecbea
Contrib modules we can defer and prioritise as more users require. I think it'd be good to include ONNX support for completion.
I'd like to get involved in updating this project.
Could a list of known tasks be compiled here please.