8f2780f579
Make BaggingClassifier use if_delegate_has_method in decision_function
2015-02-26 10:53:37 +01:00
0036dd0319
ENH make defensive copies in GradientBoosting*.staged_decision function.
2015-02-24 16:28:33 -05:00
92ed5eee42
FIX ensure that negative float max_features will lead to an error
2015-02-03 13:39:18 +01:00
92b90815a6
FIX Ensure at least 1 feature is sampled when max_features is a float
2015-02-03 13:30:17 +01:00
35c2535bfc
rename cw option to subsample & refactor its implementation
2015-01-12 21:27:13 -08:00
b541191754
Merge branch 'master' into rf-class_weight
...
Conflicts:
doc/whats_new.rst
sklearn/ensemble/forest.py
sklearn/tree/tree.py
2015-01-12 19:15:21 -08:00
0bca2ffdfa
FIX Refactor _is_fitted method into check_is_fitted.
...
DOC Make the docstring of NotFittedError class to conform to pep8.
DOC Make the error message of check_is_fitted more newbie friendly.
2015-01-12 00:33:40 +05:30
129e319f99
FIX the try except block to recognize NotFittedError
2015-01-12 00:33:40 +05:30
9c7566e5df
Merge branch 'master' into rf-class_weight
...
Conflicts:
doc/whats_new.rst
2015-01-08 07:56:19 -08:00
2b24ccefa0
parameter validation checks & tests for errors
2015-01-03 13:15:42 -08:00
085d677ea2
add class_weight to trees, expand tests & minor refactor
2014-12-22 16:50:58 -08:00
5b90e97ebf
TST remove SVR(probability=True) from AdaBoost tests
...
Doesn't seem to have been testing anything, because CustomSVR doesn't
have predict_proba.
2014-12-17 10:37:33 +01:00
b131ad23fa
add multioutput support & bootstrap auto mode
2014-12-14 13:31:31 -08:00
de35beab43
add support for class_weights
2014-12-11 16:22:26 -08:00
8621f00f6f
Merge pull request #3173 from arjoly/sparse-tree
...
[MRG] Sparse input support for decision tree and forest
2014-11-21 10:10:33 +01:00
ee53f51f15
TST stronger non-regression test for #3815
...
Also fixed the testing imports.
2014-11-17 23:00:33 +01:00
248a5037f6
Fixed issue 3815. Discrete AdaBoostClassifier now fails early if the base classifier if worse than random.
2014-11-07 13:42:19 +00:00
c03c01a504
ENH Bring sparse input support to tree-based methods
...
Author: Arnaud Joly <arnaud.v.joly@gmail.com>
Fares Hedayati <fares.hedayati@gmail.com>
2014-11-06 13:58:28 +01:00
6b548f96f1
MAINT remove deprecated loss in gradient boosting
2014-10-24 12:58:13 +02:00
d4405cbf6c
MAINT remove deprecated oob_score_
2014-10-21 17:33:48 +02:00
cd7b43ccf9
TST make catch_warnings blocks more robust
2014-10-12 18:20:18 +02:00
d7c17a1ff5
TST bagging of pipeline of classifier
2014-10-07 10:50:36 +02:00
4d94392b49
Added _check_sample_weight
2014-09-28 19:23:36 +05:30
16bc265ed8
dded tests to test_boosting
2014-09-28 14:58:55 +05:30
85b2c783ce
add probability output for exponential loss
2014-09-22 00:29:38 +02:00
e6fa80014b
non-uniform weight test case for both reg and clf and all losses
2014-09-22 00:12:38 +02:00
05fbc6423d
fix: sample_weight is None in Huber.deviance
...
fix: sample_weight multiplication in Huber leaf updates
more tests (non-uniform weights, weight consistency and invariance to scaling, deviance consistence)
2014-09-18 10:42:36 +02:00
75f6e015c9
more tests for weighted percentile
2014-09-17 12:14:26 +02:00
d9925961df
fix: add **kwargs to Multinomial loss' negative_gradient
2014-09-17 12:14:26 +02:00
93b2b7dae1
fix: consider sample_weights in robost init estimator and negative_gradient
2014-09-17 12:14:26 +02:00
fd952cc0ec
sample weight support for robust regression via weighted percentile algo
2014-09-17 12:14:26 +02:00
297b36a9b1
update tests for new exception types
2014-09-17 12:14:26 +02:00
6fe6c34c15
adressed Olivier's feedback
2014-09-17 12:14:26 +02:00
2503d35039
cosmit: doc
...
raise ValueError for sample_weights and robust regression
2014-09-17 12:14:26 +02:00
f98388e31f
more tests
2014-09-17 12:14:26 +02:00
7f4c974d6b
Sample weights for gradient boosting
...
Exponential loss for classification + predict_proba refactoring
2014-09-17 12:14:26 +02:00
aeb94c647d
ENH: parallel feature importances for forests
2014-08-06 11:40:46 +02:00
a7c9128080
ENH removed manual code for parallel task batching forests
...
Forests now use the threading backend that has such a low overhead
that batching tasks together does not bring any measurable
performance benefit.
This removes the boilerplate to simplify the code and make it
easier to understand and maintain.
Furthermore, batching will soon be implemented in joblib. So even for
models that use the multiprocessing backend, manually batching tasks
will be useless at some point.
2014-08-06 11:40:45 +02:00
ffd13c389b
fix: i might be undefined
2014-07-31 23:47:51 +02:00
9131413aaa
Added warm start to random forests.
2014-07-25 13:58:06 +02:00
c9789c7edf
COSMIT clean up tests with pyflakes
...
test_ridge.py contained two duplicate functions.
2014-07-22 16:10:16 +02:00
4113dfe12d
Added a parameter check on n_estimators in BaseEnsemble to raise error if it is not strictly positive.
2014-07-18 19:26:33 +02:00
46c1c0ac82
MAINT faster test_weight_boosting
2014-07-17 17:09:21 +02:00
32092817bd
ENH improve forest testing + avoid *args
2014-07-16 15:59:34 +02:00
79d965afcc
min_weight_fraction: comments and pep8
2014-07-02 11:47:32 +02:00
987170713e
forest: test min_samples_leaf and min_weight_fraction_leaf
2014-07-02 11:47:32 +02:00
72c60eba94
gbrt: min_weight_fraction_leaf value test
2014-07-02 11:47:31 +02:00
c7db70ac9f
removed unused variables and imports, add used imports where missing.
2014-06-08 21:44:01 +02:00
32f006914f
Fixed paramaters to make_multilabel_classification and make_regression to keep tests under 1 second, reformated some blank lines
2014-06-03 11:25:59 -04:00
bb752bd70e
Remove exectuable permision bit
2014-06-02 20:30:38 -04:00
Andreas Mueller