add lightgbm.booster support #270

qh582 · 2018-05-29T03:55:41Z

mainly followed the logic of xgboost.booster support

add lightgbm.booster support

lgb2.1.0 to lgb 2.1.1

update to lgb2.1.1

codecov-io · 2018-05-29T04:49:49Z

Codecov Report

Merging #270 into master will decrease coverage by <.01%.
The diff coverage is 97.91%.

@@            Coverage Diff             @@
##           master     #270      +/-   ##
==========================================
- Coverage   97.22%   97.22%   -0.01%     
==========================================
  Files          44       44              
  Lines        2815     2845      +30     
  Branches      536      545       +9     
==========================================
+ Hits         2737     2766      +29     
  Misses         41       41              
- Partials       37       38       +1

Impacted Files	Coverage Δ
eli5/lightgbm.py	`95.23% <97.91%> (+0.36%)`	⬆️

lopuhin · 2018-05-29T07:23:03Z

eli5/lightgbm.py

        proba=proba,
        get_score_weights=get_score_weights,
     )

-
+def _check_booster_args(lgb, is_regression=None):
+    # type: (Any, bool) -> Tuple[Booster, bool]


I hope something like # type: (Any, bool) -> Tuple[lightgbm.Booster, bool] and adding Tuple and Any to typing imports at the top should fix the build (https://travis-ci.org/TeamHG-Memex/eli5/jobs/385021206#L764-L766). To check locally, you can try installing mypy==0.550 and running mypy --check-untyped-defs eli5

lopuhin · 2018-05-29T07:26:31Z

Thanks for the PR @qh582 , at first sight it looks good 👍
I left a comment on how to fix the build. It would be also great if you could add tests for lightgbm.booster support, so that we don't break it in the future.

qh582 · 2018-05-30T03:36:37Z

@lopuhin Thanks for review and suggestion.

solved it.

I am doing the explain_prediction for regression part test, and I got 'y=y' in stead of 'y' as shown in the screenshot:

This seems ok for me, but I found eli5 using assert '<b>y</b>' in strip_blanks(expl_html) in tests.test_sklearn_explain_prediction.assert_trained_linear_regression_explained which point that it shoud be 'y' in my case. I am wondering if I missed the point here.

add test

add explain_prediction test for regression booster

fix bug of using multiclass booster to fit binary case

qh582 · 2018-05-31T05:19:47Z

@lopuhin could you please check it?

lopuhin

Looks great, thanks @qh582 ! I'm +1 to merge this, @kmike do you want to have a look?

adding test done

kmike · 2018-05-31T11:03:01Z

@qh582 looks good! Would you mind updating a few notes in docs, mentioning this new feature?

* Update overview.rst * Update README.rst * Update lightgbm.rst

qh582 · 2018-05-31T12:17:05Z

@kmike I think this is ready, please check :)

kmike · 2018-05-31T12:28:31Z

docs/source/libraries/lightgbm.rst

+  ("objective" starts with "reg")
+  and False for a classification problem.
+  If not set, regression is assumed for a single target estimator
+  and proba will not be shown unless the ``target_names`` is defined as a list with length of two.


This seems to contradict a note above (line 25, "target_names is ignored") - does the note about target_names apply only to classifier/regressor, but not for Booster?

for a single target estimater, booster cannot recognize wheather it is a regression problem or not. We assume it is regression in default, unless users set it as a classification problem by assigning 'target names' input [0,1] etc. Only in this case 'target names' is used. Should I remove " target names is ignored" to prevent confusion？

target names is still ignored for classifier/regressor, right? I think it makes sense to clarify it - just say that it is ignored for LGBMClassifier / LGBMRegressor, but used for lightgbm.Booster.

"is defined as a list with length of two" - sohuld it be 2 elements, or 2+ is also supported?

Well, for single target booster, only 1+ elements target_name point booster to classification. Otherwise, this booster is regression . there is risk user assign wrong number of elements like 3, but not sure how eli5 will raise error. 2+ should not support here.

kmike · 2018-05-31T12:29:32Z

docs/source/libraries/lightgbm.rst

-and :func:`eli5.explain_prediction` for ``lightgbm.LGBMClassifer``
-and ``lightgbm.LGBMRegressor`` estimators.
+and :func:`eli5.explain_prediction` for ``lightgbm.LGBMClassifer``, ``lightgbm.LGBMRegressor`` and ``lightgbm.Booster``  estimators. It is tested against LightGBM
+master branch.


"It is tested against LightGBM master branch." is removed in 63e9918, sorry for the confusion. It shouldn't be re-added.

got it, I will remove that line in next commit

qh582 · 2018-05-31T15:10:28Z

@kmike Hope the new commit help. My previous piece of code is not strict enough.

WestXu · 2018-08-08T06:12:05Z

It's been two month, could someone tell me why this merge hasn't been closed already? I really need this.

lopuhin · 2018-11-14T17:59:46Z

I think PR is in great shape and we just need to check it again and merge it, hope to find time for it soonish, sorry for delay.

kmike · 2018-11-18T09:49:53Z

docs/source/libraries/lightgbm.rst


 .. _LightGBM: https://github.com/Microsoft/LightGBM

 :func:`eli5.explain_weights` uses feature importances. Additional
-arguments for LGBMClassifier and LGBMClassifier:
+arguments for LGBMClassifier , LGBMClassifier and lightgbm.Booster:


a nitpick: extra whitespace before comma

kmike · 2018-11-18T09:50:30Z

docs/source/libraries/lightgbm.rst

@@ -22,7 +21,7 @@ arguments for LGBMClassifier and LGBMClassifier:
  - 'weight' - the same as 'split', for better compatibility with
    :ref:`library-xgboost`.

-``target_names`` and ``target`` arguments are ignored.
+``target_names`` arguement is ignored for ``lightgbm.LGBMClassifer`` / ``lightgbm.LGBMRegressor``, but used for ``lightgbm.Booster``. ``target`` argument is ignored.


Suggested change

``target_names`` arguement is ignored for ``lightgbm.LGBMClassifer`` / ``lightgbm.LGBMRegressor``, but used for ``lightgbm.Booster``. ``target`` argument is ignored.

``target_names`` argument is ignored for ``lightgbm.LGBMClassifer`` / ``lightgbm.LGBMRegressor``, but used for ``lightgbm.Booster``. ``targets`` argument is ignored.

kmike · 2018-11-18T09:50:42Z

docs/source/libraries/lightgbm.rst


 .. _LightGBM: https://github.com/Microsoft/LightGBM

 :func:`eli5.explain_weights` uses feature importances. Additional
-arguments for LGBMClassifier and LGBMClassifier:
+arguments for LGBMClassifier , LGBMClassifier and lightgbm.Booster:


Suggested change

arguments for LGBMClassifier , LGBMClassifier and lightgbm.Booster:

arguments for LGBMClassifier, LGBMClassifier and lightgbm.Booster:

kmike · 2018-11-18T09:51:12Z

docs/source/libraries/lightgbm.rst

@@ -22,7 +21,7 @@ arguments for LGBMClassifier and LGBMClassifier:
  - 'weight' - the same as 'split', for better compatibility with
    :ref:`library-xgboost`.

-``target_names`` and ``target`` arguments are ignored.
+``target_names`` arguement is ignored for ``lightgbm.LGBMClassifer`` / ``lightgbm.LGBMRegressor``, but used for ``lightgbm.Booster``. ``target`` argument is ignored.

 .. note::
    Top-level :func:`eli5.explain_weights` calls are dispatched


I think lightgbm.Booster should be mentioned here as well.

kmike · 2018-11-18T09:52:55Z

docs/source/libraries/lightgbm.rst

@@ -50,6 +49,14 @@ for ``lightgbm.LGBMClassifer`` and ``lightgbm.LGBMRegressor``:
  estimator. Set it to True if you're passing ``vec``,
  but ``doc`` is already vectorized.

+``lightgbm.Booster`` estimator accepts one more optional argument:
+
+* ``is_regression`` - True if solving a regression problem


Is this parameter supported? It is not an argument of explain_prediction_lightgbm.

kmike · 2018-11-18T09:53:10Z

docs/source/overview.rst

@@ -31,7 +31,7 @@ following machine learning frameworks and packages:
  of XGBClassifier, XGBRegressor and xgboost.Booster.

 * :ref:`library-lightgbm` - show feature importances and explain predictions
-  of LGBMClassifier and LGBMRegressor.
+  of LGBMClassifier , LGBMRegressor and lightgbm.Booster.


Suggested change

of LGBMClassifier , LGBMRegressor and lightgbm.Booster.

of LGBMClassifier, LGBMRegressor and lightgbm.Booster.

kmike · 2018-11-18T09:56:16Z

eli5/lightgbm.py

-    ``target_names`` and ``targets`` parameters are ignored.
+    ``target_names`` arguement is ignored for ``lightgbm.LGBMClassifer`` / ``lightgbm.LGBMRegressor``, 
+    but used for ``lightgbm.Booster``. 
+    ``target`` argument is ignored.


Suggested change

``target`` argument is ignored.

``targets`` argument is ignored.

Dronablo · 2020-04-11T19:21:33Z

Any plans about merging it, @lopuhin?

lopuhin · 2021-02-10T07:55:04Z

Thanks you! This were merged in eli5-org/eli5#7 and released to PyPI with v0.11

qh582 added 4 commits May 29, 2018 11:34

add lightgbm.booster support

d369977

mainly followed the logic of xgboost.booster support

Merge pull request #1 from qh582/qh582-lightgbm-booster

cc5b7a4

add lightgbm.booster support

fix bug

9c2bb8e

lgb2.1.0 to lgb 2.1.1

bug fix

9c31a99

update to lgb2.1.1

lopuhin reviewed May 29, 2018

View reviewed changes

qh582 added 3 commits May 29, 2018 19:20

fix build for typing

a042650

add _check_booster_args test

72a50a2

add explain_weights test for booster

bb19a4f

qh582 added 9 commits May 30, 2018 11:54

Merge pull request #2 from qh582/qh582-lightgbm-booster

4bf2f6d

add test

correct is_regression

ef991cf

add explain_prediction test for regression booster

45379f5

Merge pull request #3 from qh582/qh582-test

a35e5c8

add explain_prediction test for regression booster

add booster multiclass test

0bbf3e2

fix booster binary case

fa1eccd

add test_explain_prediction_booster_binary

f053399

remove redundant code

87438d4

fix bug of using multiclass booster to fit binary case

add _lgb_n_targets test

3c213e9

lopuhin approved these changes May 31, 2018

View reviewed changes

Merge pull request #4 from qh582/qh582-lightgbm-booster

866015a

adding test done

qh582 added 3 commits May 31, 2018 19:56

Update docs

802e543

* Update overview.rst * Update README.rst * Update lightgbm.rst

update docs

5155979

* Update overview.rst * Update README.rst * Update lightgbm.rst

Merge branch 'master' into qh582-lightgbm-booster

6f1ee40

kmike reviewed May 31, 2018

View reviewed changes

update booster is_regression

0a488e3

Merge branch 'master' into qh582-lightgbm-booster

19e9bf3

lopuhin mentioned this pull request Nov 14, 2018

no support for LGBMRegressor #280

Closed

kmike reviewed Nov 18, 2018

View reviewed changes

kmike mentioned this pull request Nov 18, 2018

DOC small fixes to xgboost docs #285

Merged

lopuhin mentioned this pull request Jan 23, 2021

Add lightgbm.booster support eli5-org/eli5#7

Merged

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

add lightgbm.booster support #270

add lightgbm.booster support #270

qh582 commented May 29, 2018

codecov-io commented May 29, 2018 •

edited

Loading

lopuhin May 29, 2018 •

edited

Loading

lopuhin commented May 29, 2018

qh582 commented May 30, 2018 •

edited

Loading

qh582 commented May 31, 2018

lopuhin left a comment

kmike commented May 31, 2018

qh582 commented May 31, 2018

kmike May 31, 2018

qh582 May 31, 2018

kmike May 31, 2018

qh582 May 31, 2018 •

edited

Loading

kmike May 31, 2018

qh582 May 31, 2018

qh582 commented May 31, 2018

WestXu commented Aug 8, 2018

lopuhin commented Nov 14, 2018

kmike Nov 18, 2018

kmike Nov 18, 2018 •

edited

Loading

kmike Nov 18, 2018

kmike Nov 18, 2018

kmike Nov 18, 2018

kmike Nov 18, 2018

kmike Nov 18, 2018

Dronablo commented Apr 11, 2020

lopuhin commented Feb 10, 2021

	``target_names`` arguement is ignored for ``lightgbm.LGBMClassifer`` / ``lightgbm.LGBMRegressor``, but used for ``lightgbm.Booster``. ``target`` argument is ignored.
	``target_names`` argument is ignored for ``lightgbm.LGBMClassifer`` / ``lightgbm.LGBMRegressor``, but used for ``lightgbm.Booster``. ``targets`` argument is ignored.

	arguments for LGBMClassifier , LGBMClassifier and lightgbm.Booster:
	arguments for LGBMClassifier, LGBMClassifier and lightgbm.Booster:

	of LGBMClassifier , LGBMRegressor and lightgbm.Booster.
	of LGBMClassifier, LGBMRegressor and lightgbm.Booster.

	``target`` argument is ignored.
	``targets`` argument is ignored.

add lightgbm.booster support #270

Are you sure you want to change the base?

add lightgbm.booster support #270

Conversation

qh582 commented May 29, 2018

codecov-io commented May 29, 2018 • edited Loading

Codecov Report

lopuhin May 29, 2018 • edited Loading

Choose a reason for hiding this comment

lopuhin commented May 29, 2018

qh582 commented May 30, 2018 • edited Loading

qh582 commented May 31, 2018

lopuhin left a comment

Choose a reason for hiding this comment

kmike commented May 31, 2018

qh582 commented May 31, 2018

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

qh582 May 31, 2018 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

qh582 commented May 31, 2018

WestXu commented Aug 8, 2018

lopuhin commented Nov 14, 2018

Choose a reason for hiding this comment

kmike Nov 18, 2018 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Dronablo commented Apr 11, 2020

lopuhin commented Feb 10, 2021

codecov-io commented May 29, 2018 •

edited

Loading

lopuhin May 29, 2018 •

edited

Loading

qh582 commented May 30, 2018 •

edited

Loading

qh582 May 31, 2018 •

edited

Loading

kmike Nov 18, 2018 •

edited

Loading