3f5277e835
Merge pull request #4284 from hbredin/dgpmm_convergence
...
[MRG + 1] Improve DPGMM/VBGMM convergence check
2015-02-25 09:46:04 +01:00
4a0e281510
DOC: make it clear 'thresh' should be removed in v0.18
2015-02-25 08:38:37 +01:00
2c3e5640d9
DOC cosmetics in test docstring
2015-02-24 23:35:24 +01:00
86392dbc6d
Put the error state change about underflow errors in a 'with' statement in '_covar_mstep_full'.
2015-02-24 23:35:24 +01:00
fa3ecbc96d
Converted docstring of 'test_positive_definite_covars' into a comment to avoid description ptoblems with nose (Related to issue #4250 )
2015-02-24 23:35:23 +01:00
db2eddaa02
Converted test_positive_definite_covars to a generator.
2015-02-24 23:35:23 +01:00
3e069d51ed
Clarified a comment and factorized the tests for all covariance types
2015-02-24 23:35:23 +01:00
e0536d93b5
Cosmetic changes in the doc string. Added an indentation level
2015-02-24 23:35:23 +01:00
739efd6e82
Added test to check that obtained covariance matrices are positive definite after learning a GMM
2015-02-24 23:35:23 +01:00
5c893e5280
Change the way the covariance is computed to avoid problem with not positively defined covariance matrices
2015-02-24 23:35:22 +01:00
0c00f1a1b5
ENH: better convergence check
2015-02-23 17:14:26 +01:00
8ef0b9a82f
fixes in GMM, TSNE, MDS, LSHForest, exclude SpectralEmbedding
2015-02-10 18:34:31 -05:00
e3e0827243
FIX check (and enforce) that estimators can accept different dtypes.
2015-02-10 15:37:15 -05:00
5c0c5c672c
MAINT use absolute imports in tests
...
as per the guideline in:
http://scikit-learn.org/stable/developers/#coding-guidelines
2015-02-10 13:40:31 +01:00
a3283c6d6b
FIX tests in gmm broken by previous rebase
2015-02-08 23:50:02 +01:00
0f6f48a6f2
a few changes according to comments by @ogrisel
2015-02-08 23:00:54 +01:00
371c2f672a
FIX: fix doctest
2015-02-08 22:59:46 +01:00
bdecd1f447
ENH: issue #4178 (cont.)
2015-02-08 22:59:46 +01:00
79144e4444
ENH: improve GMM convergence check #4178
2015-02-08 22:59:45 +01:00
364fdd6496
minor fixes, addressing @agramfort's comments.
2015-02-03 17:30:08 +01:00
87c9ba7d26
Corrected two bugs related to 'tied' covariance_type in mixture.GMM(), added test, closes #4036
...
The first bug was with the Gaussian log-density calculation for the 'tied' covariance_type.
Rather than fix this equation I reformatted the covars data structure so that it could be fed
into the 'full' covariance_type log-density calculation. It might be a tiny bit slower but I think
it is better to have the least amount of potentially redundant code as this should minimize
the potential for such errors.
The second bug I fixed was related to the _covar_mstep_tied() function, only the first part
of the equation should be divided by X.shape[0].
2015-02-03 17:29:26 +01:00
b3fbccca38
FIX various mismatch between docstring and signature params
...
DOC max_iterations -> max_iter. Make it consistent with kmeans
MAINT Replace the deprecated dx parameter with d in the docstrings
MAINT Deprecation warning for max_iterations parameter.
2015-01-16 11:54:57 -05:00
abd31d2cad
MAINT Make uniform the error raised for not fitted condition
2015-01-12 00:33:40 +05:30
88555987cc
PEP8 Fix E101, E111 errors and W191, W293, W293 and W391 warnings.
2015-01-08 04:28:24 +05:30
52472bc924
unused variables are redefined later
2014-11-07 10:31:46 -05:00
f0405a2a19
DOC: Remove extra # from url in fit.
...
The extra # messes with the rst parser which produces an invalid URL where the ".html" suffix is missing.
2014-10-24 12:16:55 -04:00
beb0843072
DOC: fix doc inconsistency in GMM
2014-10-18 12:46:07 -07:00
b3bdb08964
DOC fix formatting of attributes etc. in docstrings
2014-07-28 19:04:59 +10:00
56057c9630
MAINT remove deprecated code
2014-07-23 16:19:40 +02:00
795a6168b4
FIX: use clip(0) instead of abs()
...
This removes spurious tiny negatives from the output of eig.
2014-05-19 11:22:07 +02:00
c6b7e3baed
FIX numerical stability in GMM with eigh sampling
2014-05-15 15:37:33 +02:00
3716e88999
ENH faster sampling in GMMs
...
A few optimizations for sampling with full covariance matrix:
1) use eigh instead of svd, which are equivalent (up to numerical precision),
2) use broadcasting instead of matrix multiplication for the product U s^(1/2),
3) simpler construction of the random vectors.
2014-05-15 14:49:25 +02:00
248cf69e55
MAINT remove default values from private funcs in GMM
...
Removed default values for 'mean' and 'covars' arguments.
They did not have the excepted format causing the functions
to fail when called with default arguments.
2014-05-15 14:48:52 +02:00
fb795ebef0
ENH factor out squared norm helper
2014-04-08 23:02:54 +02:00
b40e58e12b
Cosmit
2014-03-23 18:46:30 +01:00
23b3cc2eb5
MAINT drop support for SciPy < 0.9
...
TODO: get rid of solve_triangular. Just replacing it by scipy.linalg's
breaks OMP and least-angle regression.
2014-03-02 20:13:57 +01:00
6ebcf4d020
COSMIT fix PEP8 errors
2014-02-02 00:30:28 +11:00
fd2226d3dc
Update dpgmm.py
...
The document has been moved.
2014-01-02 21:47:31 +08:00
dea7bbb2ed
BUG: typo fixes in sklearn.mixture.gmm
2013-11-26 15:33:53 +01:00
2eddb1a29f
BUG fix whitespace in error messages
...
Partial fix for #2380 .
Signed-off-by: Lars Buitinck <L.J.Buitinck@uva.nl>
2013-08-26 17:02:52 +02:00
32eb8b57f8
ENH rename eval / pseudolikelihood to score_samples
2013-07-26 23:32:23 +02:00
4a4c2cff59
Documentation ERROR: mixture.DPGMM.precs_
...
in the code: mixture.DPGMM.precs_
in the doc: mixture.DPGMM.precisions_
2013-06-04 07:43:17 +02:00
f1df9ab19d
Replaced 'for i' with 'for _' at place where i is not used.
...
fix pep8 for weighted_boosting.py
2013-04-27 15:24:44 +02:00
63562b0416
PY3: replaced izip
2013-03-19 18:02:34 +01:00
541987802f
P3K range vs. xrange
2013-02-14 02:05:35 +01:00
7d57b9e50e
COSMIT pep8, removing unused imports
2013-02-11 19:18:01 +01:00
1967a0b323
P3K use six to have a python 2 & 3 compatible code base
2013-02-09 18:08:44 +01:00
e369c4339d
COSMIT remove unused imports, pep8
2013-01-26 18:26:20 +01:00
07c56d7cd2
COSMIT rm deprecated stuff
...
One more deprecated feature to go in FastICA; not sure how to handle
that one.
2013-01-22 19:45:27 +01:00
47b1ba0d0e
COSMIT: pep8
2013-01-17 17:35:31 +01:00
Olivier Grisel