Timeline for Why is there no improvement when training Xgboost with pseudo-Huber loss?
Current License: CC BY-SA 4.0
11 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Nov 30, 2023 at 1:59 | comment | added | usεr11852 |
+1 fun question. A pseudohubererror objective is slightly more involved than it might seem first and can require some additional tuning, please see my answer below for more details.
|
|
| Nov 30, 2023 at 1:57 | answer | added | usεr11852 | timeline score: 4 | |
| Jan 16, 2022 at 9:00 | history | tweeted | twitter.com/StackStats/status/1482638768268099589 | ||
| Aug 31, 2021 at 16:41 | comment | added | PaulG | @wdkrnls: Sorry, I meant huber-loss and related functions such as log-cosh did not improve (log-cosh and I believe 2 other huber-loss implementiations were the only custom ones I tried). Other custom losses (not Huber related) I used like rmse, mse, earth-mover-distance-based losses etc. behaved as expected. | |
| Aug 30, 2021 at 15:04 | comment | added | wdkrnls |
That's good. I thought in the other question you had tried to use log_cosh and didn't have much luck with that either. I tried that one myself and couldn't get it to work, so you beat me there.
|
|
| Aug 29, 2021 at 8:33 | comment | added | PaulG | @wdkrnls: Yes I have, to check if the rmse corresponds to the native function - and it did as far as I remember. I have tried multiple custom objective functions and they all improved, except the Huber loss. | |
| Aug 27, 2021 at 22:30 | comment | added | wdkrnls | Have you tried implementing mean squared error as a custom objective function? | |
| Mar 21, 2021 at 13:08 | history | edited | PaulG | CC BY-SA 4.0 |
added code
|
| Mar 20, 2021 at 13:39 | history | edited | PaulG | CC BY-SA 4.0 |
added 72 characters in body
|
| Mar 20, 2021 at 12:48 | history | edited | PaulG | CC BY-SA 4.0 |
added 108 characters in body
|
| Mar 20, 2021 at 12:30 | history | asked | PaulG | CC BY-SA 4.0 |