808fa6450e
FIX consistency of memory layout for linear CD solver
2015-10-04 18:32:05 +02:00
fead69a4ac
Bugfix : type in cd changed for read only memmap compatibility
2015-07-31 12:12:40 +02:00
3a33776003
FIX : avoid unecessary computation sparse dual gap
2015-01-04 22:18:19 +01:00
150b4913ed
OPT: Prevent iterating across n_features twice
2014-09-10 10:24:20 +02:00
0535fe9b70
BUG: n_samples instead of n_features in cd_fast
...
Was causing a segfault under numpy master, that is more sensitive to
array index overflow. Our code was wrong, numpy is not to be blamed.
2014-08-31 11:55:27 +02:00
cf4cf60b7e
ENH Stochastic Coordinate Descent for ElasticNet & Lasso
2014-07-29 10:23:33 +02:00
947da72a0a
TST: Added test to check that warm model converges faster than a cold one
2014-07-15 12:17:36 +02:00
d7f4142728
ENH: Return attribute n_iter_ for linear models dependent on the enet solver
2014-07-15 12:17:35 +02:00
2c83678022
ENH: Replaced double indexing as it causes python overhead
2014-06-13 16:41:11 +05:30
75b48cd5b2
Release GIL in the multi-task variant
2014-06-12 12:46:54 +05:30
70d3cce5c8
Added typed memory-views for H and XtA
2014-06-11 17:25:03 +05:30
d49722037e
Replaced C ordered contiguity and added typed memory views
2014-06-10 22:07:10 +05:30
3e6033106b
ENH: Release GIL in the gram variant
2014-06-10 21:05:08 +05:30
acd5df39df
COSMIT: Replaced xrange with range
2014-06-10 21:05:08 +05:30
5a839e6f70
ENH: Release the GIL for sparse coordinate descent
2014-06-10 21:05:08 +05:30
7cf73f4151
Made the following fixes
...
1. Replaced l1_norm with dasum calls
2. Changed strides to n_tasks in ddot
2014-06-10 21:05:08 +05:30
5d68c27815
Replaced numpy calls with blas in dual_gap
2014-06-10 21:05:08 +05:30
89296efa50
WIP: releasing the GIL in the inner loop of coordinate descent
2014-06-10 21:05:07 +05:30
a786e12ace
TST/COSMIT tests/comments for [sparse]center_data
...
Also removes sparse_std which was redundant
2014-03-30 22:07:48 +11:00
878cd8b8a4
Speed up using typed memory views
2014-03-13 00:07:58 +05:30
b911a5e4a9
Replaced cython calls for dot operations with BLAS calls
2014-03-12 01:02:39 +05:30
26d3485d00
ENH: Speed up sparse_coordinate descent
2014-03-11 19:31:59 +05:30
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
37c1d092f8
COSMIT shut up the build by calling np.import_array in Cython modules
...
$(make in 2>&1 | wc -l) down from 706 to 517.
Took the liberty of modernizing some of the Cython code while I was at it.
2013-01-10 18:00:21 +01:00
0984a2a59f
cd_fast: use square norm directly
...
Previously the ℓ₂ norm was computed (which uses a sqrt), then squared.
Also, ℓ₂ norm is computed using np.dot.
2012-10-03 19:12:47 -07:00
86e36168ec
FIX : remove unnecessary dgemm in cd_fast.pyx
2012-08-07 15:51:13 +02:00
21706e36b9
ENH + DOC : add MultiTaskElasticNet + doc + 1 example
2012-08-07 15:51:12 +02:00
dcbf5a61d2
ENH : MultiTaskLasso works (still draft)
2012-08-07 15:51:12 +02:00
91f1013965
fix pip8
2012-06-09 11:13:06 +02:00
096e3e0bf2
add sparse_std that has been wrongly removed in commit 48ba97f1 from the
...
sparse version
2012-06-08 14:11:07 +02:00
91a5b42f57
merge cd_fast and cd_fast_sparse
2012-06-07 16:21:20 +02:00
1926152715
ENH : add duality gap check with Lasso(positive=True)
2012-04-10 13:59:02 +02:00
a7215af854
- changed parameter from positive_constraint to positive
...
- moved parameter from fit to constructor
2012-03-31 23:49:58 +02:00
8fbdde0ca3
renaming
2012-03-31 20:41:27 +02:00
76f34bbc35
remove outcommented lines
2012-03-31 20:18:54 +02:00
a78715ad89
positive constrained option for lasso added
2012-03-31 20:04:32 +02:00
10f18e6655
Revert "ENH: small speedups in coordinate descent"
...
This reverts commit d434a39e76
2012-02-25 19:18:01 +01:00
d434a39e76
ENH: small speedups in coordinate descent
...
speedup of only ~ 5%, but for little cost
2012-02-25 19:04:21 +01:00
4d652d1f97
MISC added warning to coordinate descent if alpha=0, don't call cd with alpha=0 in tests.
2012-02-11 16:07:07 +01:00
ddf4b72109
Move project directory from scikits.learn to sklearn
2011-09-02 12:06:57 +02:00
68f27e3790
Revert "Move project directory from scikits.learn to sklearn"
...
This reverts commit fd0d3b879d
2011-09-02 12:03:18 +02:00
fd0d3b879d
Move project directory from scikits.learn to sklearn
2011-09-02 11:38:24 +02:00
Olivier Grisel