b2d6a30aeb
IncrementalPCA implementation
2014-09-22 13:12:29 -04:00
05e65a655d
TEST: fix test failing due to numeric instability
2014-07-14 20:47:18 +02:00
b70a481430
FIX: More robust test_extmath.py, compatible with older numpy
2014-04-17 11:09:11 -04:00
fb795ebef0
ENH factor out squared norm helper
2014-04-08 23:02:54 +02:00
eb7e86a7eb
TST don't run fast_dot tests on numpy>=1.7.2 + pep8
2014-03-30 12:04:33 +02:00
20b87f9dce
TST: test utils.extmath more robust
2014-03-02 20:13:57 +01:00
fd54575111
ENH re-instate extmath.logistic_sigmoid
...
For backward compat only, so immediately deprecated.
2014-02-23 17:47:03 +01:00
0ba1e28cd5
ENH speed up RBM training with scipy.special.expit
...
SciPy 0.10 already has an implementation of the logistic function called
scipy.special.expit.
Using this makes RBM training 12% faster, as measured by observing the time
per iteration as reported by the plot_rbm_logistic_classification.py example,
disregarding the first iteration as an outlier (it's consistently faster than
the rest, not sure why). The speedup is in the expit function, not the
inplace operations; those are there to make benchmarking easier, but they
certainly won't hurt.
2014-02-23 17:46:53 +01:00
dbc26ce996
COSMIT remove unused imports and variables
2014-02-03 19:43:56 +11:00
3a6c0344e9
ENH: more of this
2013-11-29 13:55:39 +01:00
d2f3a8a3da
ENH: refactor warnings 1
2013-11-29 13:55:38 +01:00
5055836043
STY: tuple for Gael
2013-11-29 13:55:38 +01:00
4d01a77a69
FIX: unintended assignment to ValueError
2013-11-29 13:55:38 +01:00
642cbb2a69
MAINT: use stable blas getter API
2013-11-25 19:16:58 +01:00
9162dd5018
More robust unit tests for fast_dot.
2013-11-09 18:21:50 +09:00
8143e40656
BUG: ValueError was assigned as local variable!
2013-11-09 13:40:09 +09:00
311172e31a
TST fix still broken test_fast_dot
...
On Travis, SciPy seems to raise AttributeError instead of ValueError.
2013-10-20 20:45:57 +02:00
c14a619de6
TST fix broken fast_dot test
...
Due to a misspelling ("get_blas" vs. "get_blas_funcs") and a bare except
clause, the tests weren't actually being run.
2013-10-20 19:17:26 +02:00
24af95a211
ENH refactor squared-norms computation to extmath
...
k-means can now use the memory-efficient dense implementation from
metrics.pairwise, while pairwise can use the fast sparse implementation
from k-means.
2013-10-12 20:27:51 +02:00
b0de2e3f5b
fix visual indentation
2013-08-31 12:28:28 +05:30
f91fba2f01
ENH: restrict fast_dot to np < 1.8
2013-08-25 11:16:22 +02:00
1322da977c
ENH/FIX: address @ogrisel comments
2013-08-25 11:16:21 +02:00
b86f3d5208
ENH: add new Warning class, improve tests, update docs
2013-08-25 11:16:21 +02:00
69fc2970e7
FIX: remove spurious test
2013-08-25 11:16:21 +02:00
5a3a432a79
ENH: return np.dot if ndim does not match
2013-08-25 11:16:21 +02:00
e98d1ccf42
FIX: workaround for missing BLAS
2013-08-25 11:16:21 +02:00
67ceee1137
COSMITS
2013-08-25 11:16:20 +02:00
bf1655104e
better handling of warning in test
2013-08-25 11:16:20 +02:00
eeaa6e1634
FIX: fix test warning
2013-08-25 11:16:20 +02:00
3f3fb83c3d
FIX: avoid zero-dvision warnings
2013-08-25 11:16:20 +02:00
19e0401c12
employ DataConversionWarning
2013-08-25 11:16:20 +02:00
2240e44f31
ENH: add fast_dot function
2013-08-25 11:16:20 +02:00
74ec752e3c
ENH speed up logistic_sigmoid (using less code)
...
Timings, before:
>>> x = np.linspace(-1000, 1000, 10000)
>>> %timeit logistic_sigmoid(x)
1000 loops, best of 3: 570 us per loop
After:
>>> %timeit logistic_sigmoid(x)
1000 loops, best of 3: 286 us per loop
For comparison:
>>> %timeit np.tanh(x)
10000 loops, best of 3: 117 us per loop
>>> %timeit .5 * (1 + np.tanh(x/2.))
1000 loops, best of 3: 301 us per loop
TODO: optimize log_logistic_sigmoid.
2013-07-28 16:44:13 +02:00
e488649af4
ENH Add fast and stable logistic sigmoid to utils and RBM
2013-07-24 17:27:22 +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
541987802f
P3K range vs. xrange
2013-02-14 02:05:35 +01:00
5b948bbcf3
TEST flipped svd remains correct
2012-12-24 15:04:36 +01:00
641d0a7e5b
Cannot easily ensure non-uniqueness without the fix, just test uniqueness
2012-12-24 15:03:49 +01:00
5aed364db5
svd_flip test fails on Travis. Change random seed, see if it helps
2012-12-24 15:03:49 +01:00
160648d8cc
Make sure svd_flip test actually tests something
2012-12-24 15:03:49 +01:00
0755f156f8
Add sign flip as flag in randomized_svd
2012-12-24 15:03:49 +01:00
9ca8f7e60a
Test more seeds for svd sign flipping
2012-12-24 15:02:24 +01:00
0df8fb20ba
Add failing test and no-op flip
2012-12-24 15:02:24 +01:00
754c2d9533
COSMIT pep8
2012-12-22 16:48:17 +01:00
6f08ae6760
added test for cartesian
2012-11-10 13:48:35 +01:00
3c854a2fef
COSMIT pep8
2012-10-31 15:36:03 +00:00
a2cd5fa94c
Fix density utility when input is sparse.
2012-10-19 16:53:36 +09:00
e69acb1619
Add common assertions to sklearn.utils.testing.
...
The goal is to have all assertions conveniently in one place.
2012-10-19 16:43:19 +09:00
83bfb4a467
Merge exthmath tests into the same file.
2012-10-19 16:31:01 +09:00
Kyle Kastner