6722d79c1d
typo: interation/iteration + re-Cythonize cd_fast.pyx
2011-05-22 20:26:47 +02:00
d85aba65fd
cosmit: pep8
2011-05-04 20:14:54 +02:00
422d16277d
cosmit: removed trailing whitespace
2011-05-04 17:34:16 +02:00
89b396f0a1
added learning rate arguments to docstrings.
2011-05-04 17:22:52 +02:00
db8fb49395
Fixed doc-tests and added doc strings for SGD learning rates.
2011-05-04 17:20:53 +02:00
326a81aaea
Added SGD regression benchmark
2011-05-04 17:20:44 +02:00
2df3f2a271
SGD module supports two additional learning rates: constant and inverse scaling.
2011-05-04 17:20:17 +02:00
6ee9d68250
rename rng to random_state
2011-05-04 09:45:16 +02:00
21dceb2605
removed old version of min_C
2011-04-17 18:22:12 +02:00
11a8110377
got rid of np.where
2011-04-17 16:01:49 +02:00
b21a30b2f0
docstrings updated
2011-04-13 11:57:20 +02:00
c8d5baa7b7
removed float64 and int32 conversion
...
... since we are not calling any native code
2011-04-13 11:53:32 +02:00
bda15f53ee
s / shape = / .reshape() /
2011-04-12 22:30:54 +02:00
e034f09dd4
min_C: fixes to the doc strings
2011-04-12 15:38:13 +02:00
7cc9673be6
min_C: let's avoid scipy.sparse top level import
2011-04-12 11:42:01 +02:00
d58cddd237
min_C: check for ill-posed problem _y * X == 0
2011-04-12 11:24:12 +02:00
0c86e6b51e
min_C: reworked _y calculation
2011-04-12 11:21:53 +02:00
b2b8c2ba6b
pep8 compliance
2011-04-12 10:58:43 +02:00
b07bdee9ef
numpy/scipy idioms as suggested by agramfort
2011-04-12 10:49:08 +02:00
91de1c2a9d
review of min_C doc strings
2011-04-11 22:07:58 +02:00
a0085eb042
l1 logreg (liblinear): minimum C (sparse version)
2011-04-11 22:07:58 +02:00
10458f6ee0
DOC: module entries in reference documentation
...
Amongst other things, this is useful for linking to these modules.
2011-04-11 22:07:58 +02:00
0c5c87ae73
Update the docstring to reflect the package name changes.
2011-04-07 12:29:36 +02:00
138e688ea6
Refactoring in svm module.
...
Renaming and module denesting. This is just a refactoring of low-level
routines to ease usage of the low-level API. Higher level API was not
changed.
2011-03-30 16:06:49 +02:00
3db9bbef5c
pep8 + oliviers remarks
2011-03-21 09:25:42 +01:00
bc16f01e24
Additional tests for sample weights.
...
Fixed some typos in doc strings.
Changed signature of coef_ for SGDClassifier; now adheres to svm convention that coef_ is shape [1, n_features] in the case of binary classification.
2011-03-15 20:50:44 +01:00
618a1b52e7
refactored SGD module (removed code duplication, better variable naming).
...
added interface for sample weight.
added seed to interface of sparse SGD classifier and regressor.
2011-03-15 16:29:49 +01:00
40f7ba34bd
ENH: Fixed seed for shuffling in SGD
...
Non reproducible random shuffling introduces too much noise in
cross-validation that makes parameter tuning hard.
2011-03-02 15:40:31 +01:00
09940a99a6
API: SVMs: eps -> tol
...
Change the eps argument to 'tol' and expose it in base classes.
The reasonning behind calling this argument tol rather than eps is that
it is a bound on the tolerance of the optimization, and it does not
relate to the machine precision (eps).
2011-03-01 17:51:50 +01:00
f79b27d9f4
Rename SparseTransformerMixin to CoefSelectTransformerMixin.
2011-02-15 17:33:34 +09:00
1fa7336b7c
Use sparse.base.SparseTransformerMixin.
2011-02-14 16:45:49 +09:00
384824906a
Add SparseTransformerMixin.
2011-02-14 16:22:41 +09:00
19bca16d5a
FIX tests when run with scikits.learn.test()
...
scikits.learn.test() is increadibly picky with lambda functions or
staticmethods.
2010-12-17 15:32:38 +01:00
a7047ca22b
liblinear bias/intercept handling
2010-12-16 12:36:15 +01:00
7d16891664
Add notes on fluctiations of liblinear.
2010-12-14 19:33:35 +01:00
89377b979a
API : maxit replaced by max_iter everywhere
...
API : removing default computation of r2 coef in coordinate_descent (speed up +
can be accessed simply eg. lasso.score(X, y))
2010-12-11 23:14:55 -05:00
264629b978
do not precompute explained_variance_ in linear model: can be too costly: use r2_score when needed instead
2010-12-12 02:56:15 +01:00
758e5c560c
Remove redundant site.cfg parsing.
2010-12-07 15:27:31 +01:00
7fceccb5fe
Merge branch 'sgd-rename'
2010-12-03 11:28:32 +01:00
a99078063a
BUGFIX in sparse.SGDRegressor
...
added more tests for SGDRegression
2010-12-03 11:25:43 +01:00
fdc0aa01ed
Merge branch 'log_proba'
...
closing pull request :
https://github.com/scikit-learn/scikit-learn/pull/17
following ticket :
http://sourceforge.net/apps/trac/scikit-learn/ticket/157
Conflicts:
scikits/learn/svm/base.py
scikits/learn/svm/libsvm.py
2010-12-02 14:45:52 -05:00
2aa32c1af8
Merge branch 'master' into sgd-rename
2010-12-01 17:05:05 +01:00
fea5d1835f
Some love for scikits.learn.svm.
...
* Lazily import scipy.sparse when possible: this way we can add
svm.sparse to svm.__init__ without loss of performance.
* Improved docstrings for dense and sparse versions.
2010-12-01 12:25:06 +01:00
560c6c7c17
More rename in the sgd module.
...
Merge stochastic_descent into linear_model.
2010-12-01 11:38:28 +01:00
ecce971725
Initial support for weighted samples in svm module.
...
Also added test for the dense and sparse case.
2010-11-26 13:44:24 +01:00
68bc58d029
glm --> linear_model rename holocaust.
...
I also moved cd_fast.pyx from its src subdirectory to linear_models,
to match the sgd module structure.
2010-11-25 16:27:30 +01:00
Lars Buitinck