9ad829b307
Merge pull request #1914 from fabianp/ridge_svd
...
MRG - Add SVD-based solver to ridge regression.
2013-06-06 03:12:54 -07:00
38c31d434a
Fix L2 regularization order in sgd_fast
2013-05-28 20:16:49 +02:00
901e9d743a
fix incorrect reference
2013-05-22 11:25:11 +02:00
89f29ba8da
DOC Fix references to missing examples
2013-05-22 11:13:59 +02:00
8668d02902
BUG: set random state in LogisticRegression
2013-05-21 17:15:38 +02:00
9a8845e913
ENH Create FeatureSelectionMixin for shared [inverse_]transform code
...
Also rename FeatureSelectionMixin -> SelectorMixin -> _LearntSelectorMixin
And rename sklearn.feature_selection.{selector_mixin -> from_model}
2013-05-21 13:53:12 +10:00
928541c805
Fixed precompute issue (again) in ElasticNet and enet_path
...
The previous fix (41ae851643https://github.com/scikit-learn/scikit-learn/issues/1734 ) in the n_samples <= n_features case, as the value of precompute was unchanged and remained truthy.
Also, the same issue existed in enet_path, so I applied the same fix.
2013-05-13 13:35:49 +02:00
0cf8e3d06d
Use ValueError for non-existant solvers
2013-05-08 10:49:19 +02:00
9589e04fd9
FIX typo in test for RdigeCV
2013-05-07 22:52:41 +02:00
65b19245bd
fixed typo
2013-05-07 17:47:07 +02:00
735a8146f5
Use Cholesky solver by default, but use SVD as fallback
...
when the cholesky fails.
2013-05-07 17:21:43 +02:00
26a86bd991
DOC return_path argument, prettier references
2013-05-05 14:38:31 +02:00
2ffb484adb
COSMIT pep8
2013-05-04 10:43:39 +02:00
a8fe35538d
BUG: solver was not passed to computational method in Ridge object
2013-05-03 19:38:09 +02:00
daf527735a
BUG disable memory-blowing SVD for sparse input in RidgeCV
...
Fixes #1921 .
2013-05-03 17:48:39 +02:00
9c84263339
Remove unnecessary code in ridge svd
2013-05-02 11:48:37 +02:00
922df5ae5e
Add SVD-based solver to ridge regression.
...
I added a solver based on the thin SVD of X to compute the ridge
coefficients. This is much more stable than the cholesky decomposition
for singular matrices.
While in the Ridge regression the Gram matrix becomes nonsingular
because of the regularization, I've come across problems when the
regularization parameter is very small. This pull request remedies,
making the algorithm defined and stable even for the border case
alpha=0.
I took the liberty to change the default algorithm of to this one
(when there are no sample weights).
2013-05-02 11:30:18 +02:00
dacfd8bd5d
DOC: Replaced all BSD style licenses with "BSD 3 clause"
...
Replaced all BSD style licenses with "BSD 3 clause"
Not checked yet!
Removed duplicate "3 clause, 3 clause"
Removed trailing period if exists
Fixed some missed licences, still about 50 to do, but those can be automated
Think I got the last of them.
Apparently me and sed have different ideas of regex.
Found a few more
2013-04-30 08:34:46 +02:00
465e053c49
COSMIT pep8
2013-04-28 15:46:13 +02:00
341f6f770d
COSMIT: address @agramfort's comments
2013-04-25 23:15:46 +02:00
3376704917
BUG: fix sparse support in ElasticNetCV
...
DOC: fix docstring: sparse matrices are accepted
2013-04-25 23:15:38 +02:00
850a8f947a
DOC: docstring of private functions
2013-04-25 23:13:43 +02:00
cf0311cd55
ENH: improve memory usage in ElasticNetCV
2013-04-25 23:13:43 +02:00
6a489ccb22
ENH: Avoid computations in ElasticNetCV
...
Before cross-validation, a first path was computed on the full data,
which is not necessary
2013-04-25 23:13:43 +02:00
0f00baff36
stochastic_gradient: describe all losses, fix epsilon description
2013-04-25 11:40:23 -04:00
29b10e85ef
stochastic_gradient: fix mistake in _init_t docstring
2013-04-25 11:16:04 -04:00
1bf52ec373
removed class_weight parameter from RidgeClassifier.fit()
2013-04-25 07:28:24 -06:00
3df730968d
Re-add deprecated class_weight parameter.
...
Now it is present in both RidgeClassifier.fit() and
RidgeClassifierCV.fit() for consistency, and it is actually used.
2013-04-25 07:28:24 -06:00
72cce4c880
ridge regression uses compute_class_weight()
2013-04-25 07:28:23 -06:00
79190e2895
BUG Add lambda_ attribute to ARDRegression after fit
2013-04-15 22:58:41 +02:00
67ece78da4
COSMIT various typofixes
...
As suggested by codespell <https://github.com/lucasdemarchi/codespell >
Signed-off-by: Stefano Lattarini <stefano.lattarini@gmail.com>
2013-04-11 21:22:14 +02:00
eae18a3784
PY3 fix metaclasses. See #1829 .
2013-04-04 01:07:20 +06:00
972e7cffa4
BUG: ElasticNectCV choosing improper l1_ratio
...
The code was lacking good tests: it had only smoke tests. Shame on
me (I am the author).
2013-04-03 09:48:28 +02:00
a1d4d184fe
COSMIT pep8
2013-04-02 13:30:55 +02:00
01801886b2
ENH multiclass probability estimates for SGDClassifier
...
Fixes #1814 .
2013-03-30 12:33:33 +01:00
bf0c1a38bb
Merge branch 'master' into pr/1798
2013-03-25 08:41:43 +01:00
b556759a56
Merge branch 'pr/1790' -- Python 3 support from PyCon sprint
2013-03-21 13:46:20 +01:00
9b660380ff
removing accidental space
2013-03-21 03:10:52 -07:00
154007b0ac
fixing bug in linear_model.SGDClassifier for multi-class warm start
2013-03-21 03:03:38 -07:00
58b96aa9ff
DOC FIX: multi-target linear model attribute shapes
...
DOC fixes and standardization in least_angle docstrings
More standardization in linear model docs
Conflicts:
sklearn/linear_model/coordinate_descent.py
sklearn/linear_model/least_angle.py
2013-03-19 19:04:11 +00:00
155c6ea57c
PY3: print function in coordinate descent doctest
2013-03-19 18:02:34 +01:00
5a5268b89f
PY3: fix remaining cStringIO imports
2013-03-19 17:57:12 +01:00
cdb9ef2d58
COSMIT prevent a copy in randomized LR
...
Also remove some superseded Py3 compat code.
2013-03-17 12:30:17 +01:00
fd90edcfc6
DOC clarify C parameter on LogisticRegression
2013-03-16 14:06:48 +01:00
f95dfe7763
BUG SGDClassifier and friends did not forget labels_ in re-fit
...
Fixes #1703 .
2013-03-15 12:37:16 +01:00
d61f36c120
Clarify docstring in lars_path
2013-03-15 10:41:25 +01:00
21ba76d4a4
DOC some sphinx / rst fixes
2013-03-10 18:45:23 +01:00
41ae851643
PEP8 and allow non-bool truth values in CD
2013-03-10 17:51:04 +01:00
29ed168831
BUG fix compute_class_weights issue in SGD
...
Fixes #1749 .
2013-03-08 17:33:08 +01:00
bde8b7a360
Update coordinate_descent.py
...
Reverting _dense_fit to explicitly check precompute == True rather than just looking at its truthiness.
Should prevent unwanted precomputing, see https://github.com/scikit-learn/scikit-learn/issues/1734
2013-03-04 19:48:10 -08:00
Fabian Pedregosa