Metric for time #108
-
|
I was curious to find that the time in the resulting dataframe of an experiment is lower than the amount of time the experiment took (The experiment took 2+days, but the displayed seconds translate to 10hours). Is there any reason simulated_time (as described in README) would show less than the actual time? I thought simulated time would add to it as it accounts for download and upload communication cost that is not really present since all data is gathered in one machine. |
Beta Was this translation helpful? Give feedback.
Replies: 1 comment
-
|
Hi Linardos, Thank you for your interest in the FeTS Challenge! Since we will be running the validation (i.e., re-training of all methods) on a common hardware setup, you will not need to worry about whether the communication cost is normalized or not, because all participants will be on a level playing field. It is expected that there will be a difference in the simulated time and wall clock time, so no worries! Cheers, |
Beta Was this translation helpful? Give feedback.
Hi Linardos,
Thank you for your interest in the FeTS Challenge!
Since we will be running the validation (i.e., re-training of all methods) on a common hardware setup, you will not need to worry about whether the communication cost is normalized or not, because all participants will be on a level playing field.
It is expected that there will be a difference in the simulated time and wall clock time, so no worries!
Cheers,
Sarthak