New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
[BUG] Use of private sklearn methods causing issues #4669
Comments
Thanks for reporting! I thought we had eliminated all private imports of I suppose this remained under the radar since it's not a private function but a private module, so it won't be caught by a search-all for I've opened a general issue for detection and testing here: #4670 And let's deal with the specific bugs you reported in this issue, @ngupta23. |
@ngupta23, question: I'm not sure how to reproduce or test this. Do you have any insights from your users who experience problems, e.g., is there a piece of code that sporadically fails, or similar? What dependency conditions does this occur under? |
Thanks. That is the only one I am aware of right now. If I encounter anything else, I will let you know. |
@fkiraly Is this issue closed since the PR is merged? |
there are other, unresolved instances of private imports:
That's probably why it hasn't been so clear to me whether to close this or not. Maybe we should close this as it originally related to the random state and open a tracker issue for the others? |
Yes, it would be better to get this in the next release (and address others as needed later). We have now seen 4 users report this error on pycaret (were 2 when I opened it) |
sorry for that - have you checked whether the bugfix would solve it, e.g., by prerelease testing? Because if not, we need to drill deeper. Btw, I remember now: the actual reason why I kept this open is that we need confirmation that this indeed resolves your (users') issue, @ngupta23 - we have/had no means of testing as there is no clear failure case. |
Btw, @ngupta23, in the new bug reports, is there anything we could use as a reproducible failure case and a regression test? |
& FYI, 0.19.2 is scheduled for the next days if all goes well. |
No, I have been unable to do it. See pycaret/pycaret#3569 (comment) |
have you been able to do a prerelease test whether current If you're not sure how, here's how I (ab)use the CI for testing with prereleases of
Or, is it with users and not in the CI? |
Why I'm asking: there's a few days until 0.19.2, so if it doesn't fix it, we lose 2-3 weeks or so until it gets fixed (assuming the best, i.e., that we can quickly find the actual error once we have the information that it's not this). Is there a way for one of your users to install |
I have posted the request to the original posters to verify the failure by installing sktime from git pycaret/pycaret#3605 (comment) |
See pycaret/pycaret#3569 (comment) for a potential solution. |
Is this resolved with 0.19.2, @ngupta23? |
Don't know. None of the OPs have responded. |
well, that's usually a cautiously good sign if people stop complaining... |
@ngupta23, any updates? Should/can we close this? Are the issues addressed? |
Out of the three who reported this issue, a couple responded and said that the upgrade did not resolve the issue (although did not provide the new error message). pycaret/pycaret#3582 (comment) - No response In the same threads, a couple of them reported the issue as resolved once they upgraded the
I think that the upgrade will at least resolve the previously reported error in this issue, so maybe you can close this for now. If it persists, we can evaluate it and reopen another issue. |
Summary of Possible Solutions:
|
ok, thanks for the update! |
Describe the bug
The following code in sktime calls a provate module in sklearn. This is causing sporadic issues in pycaret.
https://github.com/sktime/sktime/blob/b1a6c0d68a66f779aa0dffcea8ac2fc685613715/sktime/forecasting/compose/_bagging.py#L196-L197C76
To Reproduce
I have not been able to reproduce this myself, but several users have reported this issue recently.
Expected behavior
Would expect the imports to go through. Can we somehow avoid using private sklearn modules in sktime code in this instance.
Additional context
Versions
The text was updated successfully, but these errors were encountered: