Ignite: Metrics computation during the training

[vfdev-5]

Hi @alykhantejani

Do you plan to integrate some metrics computation in the trainer or this should be done on the user side with event handlers (and logging in the same way as ignite) ?
Anyway, it would be good to display actually computed loss values.

[alykhantejani]

Hi @vfdev-5,

I plan to have metric computation available as handlers (probably under ignite.handlers.metrics) which users can then attach to events.

These can take a logger argument, which could default to print but the user can pass in my_logger.info instead for example and the handler will just call logger(result)

As always PRs are welcome :)

[alykhantejani]

@vfdev-5 some basic metrics have now been merged. Please open an issue for any specific metric request you may have

[sjsy]

@alykhantejani
Hi,
I am newbie in pythorch. Is there a simple example that uses some metrics?
Thanks

[vfdev-5]

@sjsy, yes, see here, for example.

[sjsy]

@vfdev-5 Thanks a lot!

[sjsy]

@vfdev-5 , in the example you pointed out, the metrics are attached to create_supervised_evaluator. How can I attach some metrics to create_supervised_trainer? So I can investigate training progress by attached metrics as well as the loss.
Thanks.

[vfdev-5]

@sjsy actually, you can do it like this and this. HTH