Can you provide an example how to use it using `stateful metrics`?

[sudonto]

Nice code, thanks. Could you provide an example how to display metrics as it is (no average)? Thanks

[yueyericardo]

Hi,

The Kbar's code is taken directly from tf.keras.utils.Progbar.
And it does not support stateful metrics.

I will try to add this feature when I find some time.

[sudonto]

After reading your code, I understand how to use stateful_metrics. Just pass variables you dont want to take the average to param stateful_metrics and you will get values as it is (no avg). This is my example:
Declaration:
kbar = pkbar.Kbar(target=len(data_loader), width=8, stateful_metrics=['loss','top1','top5'])

Update progbar:
kbar.update(i, values=[('batch_time', batch_time.val), ('loss', losses.avg), ('top1', top1s.avg), ('top5', top5s.avg)])

I like your work, thanks!

[yueyericardo]

After reading your code, I understand how to use stateful_metrics. Just pass variables you dont want to take the average to param stateful_metrics and you will get values as it is (no avg). This is my example:
Declaration:
kbar = pkbar.Kbar(target=len(data_loader), width=8, stateful_metrics=['loss','top1','top5'])

Update progbar:
kbar.update(i, values=[('batch_time', batch_time.val), ('loss', losses.avg), ('top1', top1s.avg), ('top5', top5s.avg)])

I like your work, thanks!

Oh, cool, I didn’t use stateful metrics before, and thought needs to modify the code.

Thanks!