You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Hi, just wondering whether you are interested in changing the ETA and AdapativeETA widgets. At the moment their estimation is based on the mean processing time per item:
pbar.seconds_elapsed / pbar.currval
and
per_item = duration / items
respectively. However, the mean is very susceptible to outliers, i.e., a few tasks that took exceptionally long will lead to over estimation of the time needed, whereas a few very short tasks (due to handled exceptions, for example) will lead to gross underestimation of the ETA. I propose to use the median instead, which is statistically much more robust. If you're interested, I can look into making a pull request.
The text was updated successfully, but these errors were encountered:
Definitely sounds interesting, although (for the sake of API stability) I would opt for either a new method or as a parameter of the ETA/AdaptiveETA methods.
If you'll make a pull request I'll definitely merge it :)
Hi, just wondering whether you are interested in changing the
ETA
andAdapativeETA
widgets. At the moment their estimation is based on the mean processing time per item:and
respectively. However, the mean is very susceptible to outliers, i.e., a few tasks that took exceptionally long will lead to over estimation of the time needed, whereas a few very short tasks (due to handled exceptions, for example) will lead to gross underestimation of the ETA. I propose to use the median instead, which is statistically much more robust. If you're interested, I can look into making a pull request.
The text was updated successfully, but these errors were encountered: