[R-package] raise an informative error when custom objective produces incorrect output (fixes #5323) #5329
Conversation
There was a problem hiding this comment.
Very nice, thanks for this!
I'm not worried about the performance overhead of doing this on every iteration, since:
length()should be O(1)- users already should expect training with custom R code to be slower than using the built-in objectives
I just left one suggestion about the language in the error messages.
R-package/R/lgb.Booster.R
Outdated
| n_hess <- length(gpair$hess) | ||
| n_preds <- length(preds) | ||
| if (n_grad != n_preds) { | ||
| stop(sprintf("grad should have length %d, got %d", n_preds, n_grad)) |
There was a problem hiding this comment.
| stop(sprintf("grad should have length %d, got %d", n_preds, n_grad)) | |
| stop(sprintf("Expected custom objective function to return grad with length %d, got %d.", n_preds, n_grad)) |
Can we please modify this error message (and the one about the hessian) to make it clearer to users that their custom objective function's implementation is the most likely cause?
There was a problem hiding this comment.
Addressed in 84e811e
|
This pull request has been automatically locked since there has not been any recent activity since it was closed. To start a new related discussion, open a new issue at https://github.com/microsoft/LightGBM/issues including a reference to this. |
Adds a check to verify that the shapes of grad and hess are correct (same as the number of predictions) to raise an informative error when they're not. This is similar to the one added in #4815 for the python-package.
Fixes #5323.