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

Document for process_type. (#7135)

Browse Source 
* Update document for prune and refresh.

* Add demo.
This commit is contained in:
Jiaming Yuan
2021-08-03 13:11:52 +08:00
committed by GitHub
parent d080b5a953
commit 7bdedacb54
4 changed files with 115 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
doc/parameter.rst
View File

@@ -177,7 +177,7 @@ Parameters for Tree Booster
- ``grow_gpu_hist``: Grow tree with GPU.
- ``sync``: synchronizes trees in all distributed nodes.
- ``refresh``: refreshes tree's statistics and/or leaf values based on the current data. Note that no random subsampling of data rows is performed.
- ``prune``: prunes the splits where loss < min_split_loss (or gamma).
- ``prune``: prunes the splits where loss < min_split_loss (or gamma) and nodes that have depth greater than ``max_depth``.
- In a distributed setting, the implicit updater sequence value would be adjusted to ``grow_histmaker,prune`` by default, and you can set ``tree_method`` as ``hist`` to use ``grow_histmaker``.

21
doc/treemethod.rst
View File

@@ -78,9 +78,24 @@ slight differences than expectation, which we are currently trying to overcome.
Other Updaters
**************
1. ``Pruner``: It prunes the built tree by ``gamma`` parameter. ``pruner`` is usually
used as part of other tree methods.
2. ``Refresh``: Refresh the statistic of built trees on a new training dataset.
1. ``Prune``: It prunes the existing trees. ``prune`` is usually used as part of other
tree methods. To use pruner independently, one needs to set the process type to update
by: ``{"process_type": "update", "updater": "prune"}``. With this set of parameters,
during trianing, XGBOost will prune the existing trees according to 2 parameters
``min_split_loss (gamma)`` and ``max_depth``.
2. ``Refresh``: Refresh the statistic of built trees on a new training dataset. Like the
pruner, To use refresh independently, one needs to set the process type to update:
``{"process_type": "update", "updater": "refresh"}``. During training, the updater will
change statistics like ``cover`` and ``weight`` according to the new training dataset.
When ``refresh_leaf`` is also set to true (default), XGBoost will update the leaf value
according to the new leaf weight, but the tree structure (split condition) itself
doesn't change.
There are examples on both training continuation (adding new trees) and using update
process on ``demo/guide-python``. Also checkout the ``process_type`` parameter in
:doc:`parameter`.
3. ``Sync``: Synchronize the tree among workers when running distributed training.
****************