Add test for xgboost modelbuiler #1359
Conversation
avolkov-intel
requested review from
Alexsandruss,
samir-nasibli and
maria-Petrova
as code owners
|
/intelci: run |
|
@razdoburdin, please, review this PR |
| @@ -0,0 +1,119 @@ | |||
| #=============================================================================== | |||
| # Copyright 2014 Intel Corporation | |||
There was a problem hiding this comment.
| # Copyright 2014 Intel Corporation | |
| # Copyright 2023 Intel Corporation |
|
@avolkov-intel - let's not use examples as mean for testing - this is foe sure doesn't worth separate example to be created |
samir-nasibli
changed the title
|
@avolkov-intel please add unit test for that feature. Let's do it on separate PR |
|
@avolkov-intel please also update doc/daal4py/model-builders.rst for this example. |
|
|
||
| # Training | ||
| xgb_clf = xgb.XGBClassifier(**params) | ||
| xgb_clf.fit(X_train, y_train, eval_set=[(X_test, y_test)]) |
There was a problem hiding this comment.
Do we really want the validation set is printed to stdout at each boosting stage?
There was a problem hiding this comment.
You should provide eval_set to fit xgboost with early_stopping_rounds (this feature is used to avoid overfitting so dataset different from train should be provided)
I suppose I will just remove example and add test in this PR since we do not need to add new example as Nikolay said |
|
/intelci: run |
avolkov-intel
changed the title
|
/intelci: run |
|
@razdoburdin, please review the PR |
|
@razdoburdin @homksei - might be it worth to just add xgboost package internally? |
it will be the best option for us |
|
/intelci: run |
tests/test_xgboost_mb.py
Outdated
| hasattr(d4p, 'gbt_classification_prediction'), | ||
| daal_check_version(((2021, 'P', 1)))]), reason) | ||
| @unittest.skipUnless(importlib.util.find_spec('xgboost') | ||
| is not None, 'xgoost library is not installed') |
There was a problem hiding this comment.
| is not None, 'xgoost library is not installed') | |
| is not None, 'xgboost library is not installed') |
razdoburdin
added
the
testing
|
/intelci: run |
|
Sorry, I see that this test passed on public CI: |
Description
In this PR #1350 fix for xgboost modelbuilder was presented. The issue was about inference of model fitted with early_stopping_rounds parameter. In this PR I present the test that shows that there is no disrepancy between xgboost and xgboost modelbuilders.