Now able to use and create custom tensorflow heads, embedders, and middleware. #135
Now able to use and create custom tensorflow heads, embedders, and middleware. #135
Conversation
…ddleware within your own packages, by allowing Parameters to have a path property.
Can one of the admins verify this patch? |
LGTM. @scttl CI is failing with @safrooze Please take a look. Note that setting a user defined Embedder/MW/Head is currently not supported with mxnet (and this PR will not fix that). |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
LGTM. Need to have tests passing, and rebase.
I couldn't find a similar area of code for mxnet to fix. If it exists I'll fix that too.
Also the CI does not appear to be passing master at the moment. It would be a bit difficult to replicate the environment on my system, but for the ones that dont have extra dependencies they pass last time I checked.
Will this be part of a versioned release?
…________________________________
From: Gal Leibovich <notifications@github.com>
Sent: Monday, December 3, 2018 7:28:01 AM
To: NervanaSystems/coach
Cc: Ryan Peach; Author
Subject: Re: [NervanaSystems/coach] Now able to use and create custom tensorflow heads, embedders, and middleware. (#135)
@galleibo-intel approved this pull request.
LGTM. Need to have tests passing, and rebase.
—
You are receiving this because you authored the thread.
Reply to this email directly, view it on GitHub<#135 (review)>, or mute the thread<https://github.com/notifications/unsubscribe-auth/AOJrye1aZvXMN4J0pgrvpGutr8s_wFsiks5u1RjRgaJpZM4Y8ksM>.
|
Yep the CI wasn't initially set to run on forked builds (just changed and kicked off a build for this and other open PRs). |
I could also run via the Amazon released docker container if that has all the dependencies. I heard about this project and am using it via the new Sagemaker RL. |
Thanks for fixing the CI much appreciated. |
What is your opinion on |
Although it differs from the solution used on other modules, I think that having it as method makes sense in this If you could rebase your branch, we can merge it to |
Ref #134
I modified the following classes:
Adding a
path
property (or function as I mention in challenges).Then I modified:
GeneralTensorFlowNetwork.get_input_embedder
GeneralTensorFlowNetwork.get_middleware
GeneralTensorFlowNetwork.get_output_head
To use these paths instead of their own local's.
I moved a local dictionary inside
GeneralTensorFlowNetwork.get_input_embedder
called mod_names toembedder_parameters.MOD_NAMES
so that it's more accessible.Challenges
InputEmbedderParameters.path
can not be a property like the rest. You can call it withemb_type
and the path will be created. But that's different than how most path's are made.Pytest
I ran pytest locally and do not see any dramatic changes in the number of passing tests.