GitLab

* Use first_metric_only flag for early_stopping function.

In order to apply early stopping with only first metric, applying first_metric_only flag for early_stopping function.

* upcate comment

* Revert "upcate comment"

This reverts commit 1e75a1a415cc16cfbe795181e148ebfe91469be4.

* added test

* fixed docstring

* cut comment and save one line

* document new feature

* always save the score of the first round in early stopping

fix #1971

* avoid using std::log on non-positive numbers

* remove unnecessary changes

* add tests

* Update test_sklearn.py

* enhanced tests

* disabled early stopping in dart mode

* check dart mode during the init phase of early_stop callback

* added docstring style test and fixed errors in existing docstrings

* hotfix

* hotfix

* fix grammar

* hotfix

* break extremely large lines in basic.py

* break extremely large lines in callback.py

* break extremely large lines in engine.py

* break extremely large lines in sklearn.py

* hotfixes

* bring consistency and clearness into early_stopping_rounds desc, metric desc and implementation

* hotfix

* hotfix

* used NDCG as default metric for lambdarank task

* fixed missed methods at ReadTheDocs and changed default eval_metric

* leaved only unique metrics

* fixed comment

The document of `early_stopping_rounds` says it will check all of
eval_set. But, this is not true. It doesn't check the dataset
specified as the training data.

This change appends an extra phrase "except the training data" to all
of the sentences "If there's more than one, will check all of them" in
documents.

* fix early stopping edge case

* fix message.

* fix tests

* fix GPU tests.

* fixed description of Dataset

* fixed description of Booster

* fixed description of train

* fixed description of cv

* fixed description of callbacks

* fixed description of plotting

* hotfixes

thanks for building in early_stopping! super useful.

* add compat.py

* unify types

* range/xrange -> range_

* move argc_ to compat.py; add is_numeric error type

* use simplejson for json

* move json to compat.py

* move pandas to compat.py

* move sklearn to compat.py

* remove unused function

* fix 'unify types'

* argc_ (lambda -> def)

* use handle is not None for _is_constructed

* sort imports; clean code; move FAQ to docs

* fix reset parameter

* redefine CVBooster

* env.model won't be None

* update env.params

* format python code with pep8

* **DO NOT MERGE** deliberately break rules to see what will happen during check

* Revert "**DO NOT MERGE** deliberately break rules to see what will happen during check"

This reverts commit 0db93cd7a43c7efa43a2112ada43d46c6f9115d9.

* fix format in test.py

* add docs for pep-8

* add troubleshooting

* use unittest

* update unittest version

* fix test_engine.py

* fix test_sklearn.py

* default eval_metric by subclass

* add test grid search

* remove verbose_eval

* add gridsearch example for python sklearn

* reset_learning_rate->reset_parameter

* make callbacks public

* add python doc

* add boosting_type to python sklearn

* fix error

current problem:
callback order: user-defined in callback parameter -> reset_learning_rate/print_evaluation/record_evaluation/early_stop
user can't insert a callback between last 4 callbacks

solution:
set order variable:
reset_learning_rate = 10
print_evaluation = 10
record_evaluation = 20
early_stop = 30
user-defined = some int
default: according to index in callback parameter list, = index - len(callbacks) (all < 0)

current callback order:
before iter: user-defined -> reset_learning_rate -> user-defined
after iter: user-defined -> print_evaluation -> user-defined -> record_evaluation -> user-defined -> early_stop -> user-defined

* add feature importances in python; add pandas support

* solve best_iteration issue

1. merge python-package
2. add dump model to json
3. fix bugs
4. clean code with pylint
5. update python examples