Hendrik/xgboost
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Revert "Fix #3485, #3540: Don't use dropout for predicting test sets" (#3563)

Browse Source 
* Revert "Fix #3485, #3540: Don't use dropout for predicting test sets (#3556)"

This reverts commit 44811f2330.

* Document behavior of predict() for DART booster

* Add notice to parameter.rst
This commit is contained in:
Philip Hyunsu Cho
2018-08-08 09:48:55 -07:00
committed by GitHub
parent e3e776bd58
commit 3c72654e3b
9 changed files with 61 additions and 30 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
doc/tutorials/dart.rst
View File

@@ -111,3 +111,9 @@ Sample Script
# make prediction
# ntree_limit must not be 0
preds = bst.predict(dtest, ntree_limit=num_round)
.. note:: Specify ``ntree_limit`` when predicting with test sets
By default, ``bst.predict()`` will perform dropouts on trees. To obtain
correct results on test sets, disable dropouts by specifying
a nonzero value for ``ntree_limit``.