Commit graph

3514 commits

Author SHA1 Message Date
Alan Saul
336f8e11c4 Added sampling for predictive quantiles and also mean and variance
where necessary
2013-10-28 15:22:06 +00:00
Alan Saul
5a924ff5cb Rederived gamma distribution 2013-10-28 15:18:43 +00:00
Max Zwiessele
067206e83e tutorial update and X observer 2013-10-28 09:14:56 +00:00
Max Zwiessele
d8151eee61 X caching is not yet done, parameter caching working fine. X cache must be adjusted to update at the right times 2013-10-27 17:04:46 +00:00
James Hensman
a889b0b7b5 fixed up plotting in sparse_gp also 2013-10-26 20:44:58 +01:00
James Hensman
eedeaa4492 fixed up the plotting 2013-10-26 19:57:21 +01:00
James Hensman
1fe92b2515 fixed up plot in GP_base 2013-10-26 15:01:35 +01:00
Max Zwiessele
d3721b76a8 changing all parameterized objects to be compatible with the new parameterization 2013-10-25 15:29:04 +01:00
Alan Saul
2fdb60287f Added derivatives for poisson and a couple of examples,
need to fix for EP.
2013-10-25 15:08:53 +01:00
Alan Saul
ba1cf96cb1 Merged with devel 2013-10-25 12:30:20 +01:00
Alan Saul
8ef3625832 Reimplemented gradients for exponential, seems to work for
laplace now, needs a visual test though
2013-10-25 12:21:11 +01:00
James Hensman
dc2a8a531e started changing the plotting in examples to remove plot_single_output 2013-10-25 09:51:41 +01:00
James Hensman
7190e0e6bb general tidying in models 2013-10-24 22:13:52 +01:00
James Hensman
eeb5f59fca improved docstrings in svigp 2013-10-24 22:06:07 +01:00
James Hensman
683f45366b some tidying in gp.py 2013-10-24 21:58:51 +01:00
James Hensman
bddb22f4af docstrings and removal of duplicated plotting code in gp_base 2013-10-24 21:30:23 +01:00
James Hensman
33b6a7d24f turned omp off by default as discussed 2013-10-24 19:32:37 +01:00
Alan Saul
a46121c430 Was a bug in the examples_tests.py, fixed and added brendan faces to ignore list 2013-10-24 15:49:20 +01:00
Alan Saul
de9e5e7fb0 Minor clean up 2013-10-24 15:21:40 +01:00
Alan Saul
9ce51e94f6 Removed unnecessary laplace examples 2013-10-24 15:19:09 +01:00
Alan Saul
8c222bef86 Updated laplace example to use predictive density aswell
as RMSE
2013-10-23 18:40:13 +01:00
Alan Saul
7b6a56f83c Added log predictive density, ln p(y*|D) 2013-10-23 18:39:48 +01:00
Alan Saul
063f747e4b Merge remote-tracking branch 'gpy_real/devel' 2013-10-23 14:41:22 +01:00
Alan Saul
3e0b597486 Updated boston tests (more folds, allow a bias as the datasets
are not normalized once split) and more folds. Tweaked some
laplace line search parameters, added basis tests for ep
2013-10-23 14:39:33 +01:00
Alan Saul
6678bca011 Fixed bug in gradient checker where it worked differently given a integer parameter to a float 2013-10-23 13:28:08 +01:00
Alan Saul
7ecf233732 Removed derivatives of variance wrt gp and derivatives of means
with respect to gp from noise models
2013-10-23 12:08:59 +01:00
Alan Saul
9b99061b09 Tore out code no longer used from noise_distributions due to
rewriting using quadrature
2013-10-22 17:31:20 +01:00
Alan Saul
c0b94f051b Added numerical mean and variance with quadrature, about to clean up 2013-10-22 17:22:23 +01:00
Max Zwiessele
e1bee4536a adjusted gaussian likelihood to new parameterization 2013-10-22 16:18:04 +01:00
Max Zwiessele
5ab7303226 gp_base newly paramterized 2013-10-22 16:17:41 +01:00
Max Zwiessele
21dd82f8d9 updated white, bias and rbf to new parameter handling 2013-10-22 16:17:13 +01:00
Max Zwiessele
27724fd456 GPRegression working, gradients still todo 2013-10-22 16:16:54 +01:00
Alan Saul
187ceeda11 Merge branch 'master' into merge_branch 2013-10-22 15:36:05 +01:00
Alan Saul
22c24c0abe Use bfgs for laplace instead 2013-10-22 15:33:14 +01:00
Alan Saul
7c9eda482c Moved transf_data to make data -1 or 1 from 0 or 1 for bernoulli
with probit into the analytical moment match (but it 10% slower),
needs removing from epmixednoise
2013-10-22 15:30:56 +01:00
Alan Saul
5f9d7eb709 Changed naming from old derivatives of likelihoods to new ones
in noise distributions
2013-10-22 15:22:27 +01:00
Alan Saul
eacf622ac7 Fixed breakage of dvar, tidied up to make more efficient 2013-10-22 13:51:16 +01:00
Max Zwiessele
0919507574 Merge branch 'devel' of github.com:SheffieldML/GPy into devel 2013-10-22 13:41:46 +01:00
Max Zwiessele
9a70f27010 first adjustments to the model and gps -> updates and gradient transforms 2013-10-22 13:41:01 +01:00
Max Zwiessele
d12f055115 parameters have a update hirarchy, in which updates to parameters are hirarchically pursuit 2013-10-22 13:39:58 +01:00
Max Zwiessele
7f05dd4a43 likelihood is now parameterized 2013-10-22 13:38:46 +01:00
Max Zwiessele
e70203a1b2 kern is now parameterized 2013-10-22 13:38:33 +01:00
Max Zwiessele
fba121c826 kern is now parameterized 2013-10-22 13:38:29 +01:00
Alan Saul
a3422eae21 Doc stringing 2013-10-22 13:37:12 +01:00
Max Zwiessele
d5afb3b797 parameters now work efficiently, tieing is iwth observer pattern 2013-10-18 16:20:01 +01:00
Alan Saul
ceb1f7490d Added quadrature numerical moment matching (but not predictive yet) 2013-10-18 16:11:47 +01:00
Alan Saul
0eee4b42d2 Fixed a few laplace bits 2013-10-18 14:08:37 +01:00
Max Zwiessele
6be05de791 still todo: untie, gradients, priors, print ties 2013-10-17 20:47:41 +01:00
Alan Saul
10f3f7d14a Refactored gradients wrt parameters slightly, need to future proof against _get_param_names() disappearing 2013-10-17 18:33:08 +01:00
Alan Saul
1848653fce Added more options to generic tests (constraining link function
values as bernoulli requies R^{0,1}) and implemented new gradients
for bernoulli
2013-10-17 17:44:08 +01:00