[Breaking] Change default evaluation metric for binary:logitraw objective to logloss #6647
Conversation
| << "' to '" << after << "'. Explicitly set eval_metric if you'd like to " | ||
| << "restore the old behavior."; | ||
| const std::string& after, const std::string& version) { | ||
| LOG(WARNING) << "Starting in XGBoost " << version << ", the default evaluation metric " |
There was a problem hiding this comment.
Note. We should plan to remove this warning soon. Probably right after 1.4.0 ?
There was a problem hiding this comment.
The warning got extra verbose when using distributed training. All workers are printing it.
There was a problem hiding this comment.
Does the warning go to logging or stdout?
There was a problem hiding this comment.
That's the _log_calback function in xgboost/core.py.
There was a problem hiding this comment.
Got it. We could modify _log_callback to use the logging module. Alternatively, we could simply remove the warning message here.
There was a problem hiding this comment.
We can do both in the future, but it will be another breaking change. To use logging, one needs to know the verbosity. So _log_callback should have 2 parameters _log_callback(msg: str, verbosity: int) to indicate the level of current message.
There was a problem hiding this comment.
Got it. Let's remove the warning message after 1.4.0 release.
|
Is there a test for the changed behaviour or a test to adapt because of the change? |
Unlike #6183, not much of the existing code is affected by this change, since |
Closes #6618
cc @lorentzenchr @mayer79