Commit graph

223 commits

Author SHA1 Message Date
mzwiessele
9a74a933f3 [updates] made updates a function 2014-09-05 12:40:20 +01:00
mzwiessele
d06f8b9272 [indexing] maybe? cannot tell, tests are broken 2014-09-04 18:02:34 +01:00
mzwiessele
4886dfde08 [indexing] maybe? cannot tell, tests are broken 2014-09-04 18:01:33 +01:00
Max Zwießele
e49bef9b94 [param] indexing fix, this can be memory intensive if millions of parameters 2014-09-04 17:30:06 +01:00
Zhenwen Dai
808cfb0501 Allow the default constraint of a Param object to be 'fixed' 2014-09-02 11:52:09 +01:00
Zhenwen Dai
1110cc31e6 some progress for parameter tie 2014-08-29 18:52:53 +01:00
Zhenwen Dai
91d1cd3131 fix the bug: the randomize function cannot properly handle variables with prior 2014-08-28 17:57:29 +01:00
Zhenwen Dai
7ed0e70a46 further bug fix for sparsegp_mpi 2014-08-27 09:45:06 +01:00
Zhenwen Dai
f29753e9e6 generalize the interface of mpi 2014-08-26 18:12:41 +01:00
Zhenwen Dai
0947112f3c Merge branch 'devel' of github.com:SheffieldML/GPy into devel 2014-08-26 16:38:04 +01:00
Zhenwen Dai
77a96efeba improve numerical stability of vardtc_parallel 2014-08-26 16:38:00 +01:00
mzwiessele
22de3ab676 [pickling] wb as write parameter 2014-08-19 10:39:13 -07:00
mzwiessele
5f863e8ca2 [parameterization] Parameter adding more robust and better error handling 2014-08-15 08:40:17 -07:00
Zhenwen Dai
785c580032 merge the current devel into psi2 2014-08-11 18:01:23 +01:00
Zhenwen Dai
6acb9b09b5 generalize the spike-and-slab prior with pi (N,Q) 2014-08-11 14:12:43 +01:00
mzwiessele
2de4b92c38 [parameter core] offset for can be done without parameter slices 2014-08-07 10:52:00 -07:00
mzwiessele
6fc982faa0 [gradcheck] some performance enhancement 2014-08-07 10:52:00 -07:00
Zhenwen Dai
98d91e6db2 Merge branch 'hmc' into devel
A HMC sampler for GP parameters
2014-08-07 14:30:03 +01:00
Zhenwen Dai
c3bb7b28a1 debug HMC shortcut 2014-08-07 14:28:46 +01:00
mzwiessele
93b92111f8 [minor] minor changes 2014-08-05 08:28:03 -07:00
mzwiessele
7c8de7fff8 [whitespaces] 2014-07-07 16:01:12 -07:00
mzwiessele
d592f680f3 logging 2014-07-02 11:15:36 -07:00
mzwiessele
89f7f2dc73 [parameterized] init greatly improved 2014-07-02 09:17:14 -07:00
Zhenwen Dai
129985998c fix ss_mrd and fix white and bias kernel 2014-07-01 13:16:57 +01:00
mzwiessele
8ef16f5529 [parameterized] bugfix: downstream parameters did not get constraint update on add_parameter 2014-06-30 09:20:56 -07:00
mzwiessele
cbcb0a9b99 [parameterized] adding parameters in hierarchy, did not update higher siblings 2014-06-30 08:18:05 -07:00
mzwiessele
8213b5011c [gp] memory > only one copy 2014-06-27 15:21:12 -07:00
mzwiessele
83e2df838d [vardtc missing] performance fixes 2014-06-27 08:04:20 -07:00
mzwiessele
66390cbf8a [optimizer array] bugfix, when updating the model the optimizer array would not update 2014-06-25 09:33:17 -07:00
Zhenwen Dai
cf33808673 fix the SSGPLVM with MPI 2014-06-25 13:57:25 +01:00
Zhenwen Dai
08ed72b2f2 fix the speed problem of the tie framework 2014-06-24 18:01:16 +01:00
Zhenwen Dai
3f36a245d1 fix the problem of multiple ties on the same param array object 2014-06-24 12:49:04 +01:00
Zhenwen Dai
567612b3a9 tie framework works roughly 2014-06-23 19:39:35 +01:00
Zhenwen Dai
fa6b5343cc merge ties branch into psi2 2014-06-23 14:37:11 +01:00
Zhenwen Dai
36d53f815c merge ties branch into psi2 2014-06-23 14:36:28 +01:00
mzwiessele
c62dd85418 [randomize] adjusted parameters to go into random generator right 2014-06-18 08:52:33 -07:00
James Hensman
0812a0e15c Revert "Minor edits to reading Lee data in datasets.py"
This reverts commit 730e229238.
2014-06-08 21:02:42 +01:00
Neil Lawrence
3003718ea9 Merge branch 'devel' of https://github.com/SheffieldML/GPy into devel 2014-06-05 18:48:55 +01:00
Neil Lawrence
730e229238 Minor edits to reading Lee data in datasets.py 2014-06-04 11:21:25 +01:00
James Hensman
ed74a81732 editied whitespace 2014-05-30 09:22:43 +01:00
Zhenwen Dai
74b62ab3c3 merge the bug of fixing function 2014-05-29 10:29:18 +01:00
James Hensman
ac2d28e2fd reverting the fixing behaviour.
two reasons: 1) the new behaviour is confusing for new users. Either
something is fixed, or it's not. 2) the fixing didn't work! things that
should have been fixed were passed to the optimizer for optimization.

If we really want to save keystrokes, consider this:

m.foo.fix()
m.foo.unfix()
m.foo.constrain_positive()

is the same as

m.foo.fix()
m.foo.constrain_positive()

but the latter throws a warning.
2014-05-29 09:55:49 +01:00
mzwiessele
3525e45b2f [copy] is now fully functional, only hierarchy observers will be copied and pickled 2014-05-23 11:11:21 +01:00
mzwiessele
9bf7e1baf6 [optimizer&fixing] optimizer has only one optimizer copy and fixing remembers old constraint 2014-05-22 12:44:50 +01:00
mzwiessele
5a2bc4863b [optimizer] one copy for the optimizer in optimizer_array, use this instead of _set|get_params_transformed 2014-05-22 11:39:04 +01:00
Max Zwiessele
43ee8ce614 [reverts] some reverts, as one param etc does not work 2014-05-21 16:49:47 +01:00
Zhenwen Dai
52c0be1848 merge devel branch in 2014-05-21 10:38:34 +01:00
Max Zwiessele
b520eb212c [fixing] fixing now saves the old constraint 2014-05-20 16:39:52 +01:00
Max Zwiessele
c507cfe4ab [index operations] added lookup for properties for a given index as dict <properties, subindex> for given index 2014-05-20 16:06:56 +01:00
Max Zwiessele
995de897b3 [core updates] first try to switch updates off and on. Use m.updates = False to switch updates off, and vice-versa 2014-05-20 14:48:53 +01:00