[
https://issues.apache.org/jira/browse/SINGA-443?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16822687#comment-16822687
]
Ngin Yun Chuan commented on SINGA-443:
--------------------------------------
Hi Liu Hui,
Sorry for the delayed response.
You can check out the documentation in
https://nginyc.github.io/rafiki/docs/latest/docs/src/user/creating-models.html.
In particular, you can the method `rafiki.model.test_model_class()` that is
supposed to help simulate a full train-inference flow in Python & check for any
errors, before submitting it to Rafiki proper.
Still, I think we could make the debugging process more efficient, ideally
establishing an efficient local model testing/debugging/development
environment, and that will be something we'll be working on in the future. In
the meantime, will appreciate if you can let us know the headaches you face
during model development right now, and we'll be sure to account for these
headaches when we roll out such a feature to assist model developers.
Thanks,
Yun Chuan
> Rafiki--How to debug rafiki efficiently.
> ----------------------------------------
>
> Key: SINGA-443
> URL: https://issues.apache.org/jira/browse/SINGA-443
> Project: Singa
> Issue Type: Wish
> Reporter: Liu Hui
> Priority: Major
>
> Recently，I try to add a model in Rafiki,such as task of image classification
> or image detection.
> I just type the code as example's, but some error came out and confused me.
> I upload the model to Rafiki, start a train job, and then check the log file
> to fix these errors, or enter a worker's docker to debug program with adding
> print.
> This method would be so inefficient.
> Are there any tutorials about how to debug Rafiki with IDE or any other
> methods which can improve efficiency of debugging.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
