Removing misleading metric log

[Vandana Kannan <va...@gmail.com>]

Hi All,

The log self.logger.info('Epoch[%d] Train-%s=%f', epoch, name, val) in python/mxnet/module/base_module.py gives the user the impression that the metric printed is for the entire epoch, which is misleading (Ref https://github.com/apache/incubator-mxnet/pull/10437). This log was maintained so that scripts that look for this text, are not broken. But this is continuing to cause confusion for users.

https://github.com/apache/incubator-mxnet/pull/12182 removes this particular log. The scripts that parse this output within incubator-mxnet have been fixed, but there is a risk that external scripts may break due to this change.

Are there any known scripts that look for this log? Any other thoughts about this PR?

Thanks,
Vandana

Re: Removing misleading metric log

[Naveen Swamy <mn...@gmail.com>]

Can you not just print the batch range[10-20] along with epoch, there might
be user scripts that depend on these values(we won't know who is using) , I
don't think we should drop this log.

On Mon, Aug 27, 2018 at 9:38 AM Vandana Kannan <va...@gmail.com> wrote:

> Hi All,
>
> The log self.logger.info('Epoch[%d] Train-%s=%f', epoch, name, val) in
> python/mxnet/module/base_module.py gives the user the impression that the
> metric printed is for the entire epoch, which is misleading (Ref
> https://github.com/apache/incubator-mxnet/pull/10437). This log was
> maintained so that scripts that look for this text, are not broken. But
> this is continuing to cause confusion for users.
>
> https://github.com/apache/incubator-mxnet/pull/12182 removes this
> particular log. The scripts that parse this output within incubator-mxnet
> have been fixed, but there is a risk that external scripts may break due to
> this change.
>
> Are there any known scripts that look for this log? Any other thoughts
> about this PR?
>
> Thanks,
> Vandana
>
>