Add tolerance to early stopping. #6942
Conversation
trivialfis
force-pushed
the
early-stop-tol
branch
from
7c86885
to
8f09468
Compare
Codecov Report
@@ Coverage Diff @@
## master #6942 +/- ##
==========================================
+ Coverage 82.01% 82.06% +0.05%
==========================================
Files 13 13
Lines 3908 3914 +6
==========================================
+ Hits 3205 3212 +7
+ Misses 703 702 -1
Continue to review full report at Codecov.
|
| """ | ||
| def __init__(self, | ||
| rounds: int, | ||
| metric_name: Optional[str] = None, | ||
| data_name: Optional[str] = None, | ||
| maximize: Optional[bool] = None, | ||
| save_best: Optional[bool] = False) -> None: | ||
| save_best: Optional[bool] = False, | ||
| tolerance: float = 0) -> None: | ||
| self.data = data_name |
There was a problem hiding this comment.
Might be worth specifying if it's absolute or relative or even providing both to be consistent with numpy etc.
There was a problem hiding this comment.
Changed the name to abs_tol with documents. I don't think relative tolerance is applicable here.
|
Can this also be added to XGBoost4J and XGBoost4J-Spark? |
|
@candalfigomoro I'm not sure, @wbo4958 has some interest in better es support in JVM packages, but that will depend on his priority. |
|
Great work! Is this also added in XGBoost on R Cran? |
Close #4982 .