Closed pshvechikov closed 11 months ago
These lines seem to be related to the smoothing algo
TB has a debiased version of what clearml has.
Thanks for highlighting the issue @pshvechikov.
We'll take a look at improving for a near release.
Hey @pshvechikov! Just letting you know that this issue has been resolved in the recently released v1.10. Let us know if there are any issues :)
@pshvechikov I'm closing the issue. Please reopen if required.
Smoothing, in the limit of its parameter, converges to a horizontal line at the value of the very first value (see image).
I would expect smoothing to work as averaging. And extreme smoothing to result in the global average of all dots (not the value of the first dot).
Is there a way to change the smoothing algorithm? Very important thing for me, as it is makes no sense to compare noisy metrics by their first values. And a lot of sense to compare by estimates of expected values.