f8d7bd43fd
ENH OneHotEncoder docs + TypeError + test active_features_
2012-11-07 17:02:56 +01:00
bf87bf5893
Merge branch 'master' into pr/1279
...
Conflicts:
doc/whats_new.rst
2012-11-07 13:36:51 +01:00
917f309559
ENH fit_transform on KMeans
...
Naive implementation for now. Fixes Issue #1157 (partly, at least).
2012-10-31 13:28:28 +01:00
427a0e6075
ENH used asarray to avoid copy
2012-10-31 08:31:24 +00:00
edcfcd3f96
COSMIT: pep8 in arrayfuncs.pyx
...
Did not rebuild the C file as the changes are only cosmetic
2012-10-31 09:28:11 +01:00
3adaaa7d4d
FIX : pass random_state to kmeans in gmm.fit
2012-10-30 11:33:23 +01:00
00d1055bf9
FIX HMM test failures
2012-10-29 17:37:12 +00:00
77b4670b6a
COSMIT use np.clip in SGD
2012-10-29 00:47:00 +01:00
0563fa9a51
BUG halve the number of LinearSVCs
...
np.random.randint can't handle the range [0..UINT_MAX] on platforms
with 32-bit long.
2012-10-28 21:27:14 +01:00
ff313267b4
FIX adjust gamma in kernelPCA tests to fit slightly modified circles with balanced classes.
2012-10-28 20:40:21 +00:00
7c0d1b686e
FIX typo in dataset generation
2012-10-28 09:04:38 +00:00
72c360ceb4
ENH simplify circles dataset generator, make classes balanced.
2012-10-28 09:04:38 +00:00
e3d369d38c
MISC: minor clean ups in hmm code
2012-10-27 23:39:54 +02:00
d5fcb73241
- check input format of MultinomialHMM.fit
...
raise ValueError in MultinominalHMM if the input does not pass the check
- pep8
2012-10-27 23:39:54 +02:00
e9f46455f5
- avoid startprob, transmat, emissionprob containing a zero element by
...
adding small number and re-normalize
- test case for above problem
2012-10-27 23:39:54 +02:00
67558036a6
pep8, ignored E126-E128
2012-10-27 23:39:54 +02:00
0ad0b3cd59
added test_fit_with_init to tests/test_hmm.py
...
also fixed a bug by this test
2012-10-27 23:39:54 +02:00
79fcb69ecc
fixed bug of initialization in hmm.py
2012-10-27 23:39:54 +02:00
0a01292dbf
BUG: restore numpy 1.3 compatbibility
2012-10-27 23:03:40 +02:00
8b6f28061d
ENH allow more than 1000 linear SVMs with custom random seeds
...
Random seed now picked in the range [0..UINT_MAX] instead of [0..1000].
Also changed the liblinear helper code to conform to C89 again, allowing
compilation with MSVC.
2012-10-27 22:55:53 +02:00
0b3bc3fadf
COSMIT: long remarks go in 'notes' section
2012-10-27 22:44:37 +02:00
a69557e3b9
TST: increase coverage in spectral clustering
2012-10-27 22:10:51 +02:00
388d1233dd
API: change 'embed_solve' to 'assign_labels'
2012-10-27 20:40:06 +02:00
0b53cb55f2
removed a hardcoded value
2012-10-27 20:40:06 +02:00
01217f3d09
Fixed some memory inefficiencies and clearned up documentation and code semantics
2012-10-27 20:40:05 +02:00
36efa068b6
Documentation changes, removed more camel case variables
2012-10-27 20:39:35 +02:00
e500133610
Even more description of the discretization algorithm.
2012-10-27 20:39:35 +02:00
40701a7a45
More description of the discretization algorithm.
2012-10-27 20:39:35 +02:00
3f00ff24d4
pep8 formatting
2012-10-27 20:39:35 +02:00
8ba410c7a5
Small aesthetic fixes to discretization
2012-10-27 20:38:41 +02:00
851a27c3c1
Added more info in documentation
2012-10-27 20:38:41 +02:00
bbfe61a254
Fix typos
2012-10-27 20:38:41 +02:00
b2502fc721
Small comments/constants added
2012-10-27 20:38:41 +02:00
c615759e59
Documentation and small bugs fixed and code cleaned up
2012-10-27 20:38:41 +02:00
7eda3998e4
Discretization method for spectral clustering added along with tolerence setting to loosen eigendecomposition constraints
2012-10-27 20:38:41 +02:00
90c007981f
BUG: make precision_recall invariant by scaling probs
2012-10-27 19:33:05 +02:00
c326b1fabe
ENH removed ``remove_zeros`` parameter.
2012-10-27 18:19:57 +01:00
402a304941
BUG: fix test error
2012-10-27 17:10:44 +02:00
d0f7da22ca
doc: added reference to lobpcg and note about small number of nodes
2012-10-27 17:06:44 +02:00
083c50715e
PEP8
2012-10-27 17:06:44 +02:00
b0f3a01a16
MISC: @agramfort's comments
2012-10-27 17:06:44 +02:00
2c857f24ee
ENH: make sure that spectral EVD is solve once
2012-10-27 17:06:44 +02:00
9a4aefdfc8
MISC: addressing @satra's comments
2012-10-27 17:06:44 +02:00
d4a6958908
FIX: numerical stability in spectral
...
This enables me to unskip a skipped test, hurray!
2012-10-27 17:06:44 +02:00
ba400b76bb
MISC: decrease verbosity by default
2012-10-27 17:06:44 +02:00
315d42111f
ENH/FIX add a lobpcg solver to spectral embedding
...
This makes the solver a bit better behaved on some problems
2012-10-27 17:06:44 +02:00
e04614ca03
TST fixed doctests, added more tests. Still 100% line coverage :)
2012-10-26 19:08:42 +01:00
f2e9d5b21f
enh masking out zero features
2012-10-26 19:46:16 +02:00
c305629f17
TST skip test in common_tests, reach 100% coverage on new code.
2012-10-25 22:57:53 +01:00
086018288d
ENH user-specified dtype, defaults to np.float, nicer numpy stuff :)
2012-10-25 22:11:53 +01:00
Lars Buitinck