Skip to content

HTTPS clone URL

Subversion checkout URL

You can clone with
or
.
Download ZIP
large-scale Gaussian process-based classification
Matlab FORTRAN C++ Other
tree: 536615e722

Fetching latest commit…

Cannot retrieve the latest commit at this time

Failed to load latest commit information.
Netlab
Selections
gpml-matlab-v3.1-2010-09-27
logB
thesis
CVP.m
CVP.m~
CV_L_LDA.m
CV_L_LDA.m~
README
README~
alg_3_3.m
alg_3_3.m~
alg_3_4.m
alg_3_4.m~
choose_subset.m
choose_subset.m~
choose_subsetP.m
choose_subsetP.m~
cross_validate.m
cross_validate.m~
cross_validate_with_learning.m
cross_validate_with_learning.m~
cv_results
del_log_det_B.m
del_log_det_B.m~
deriv_check.m
derivations.latex
derivations.pdf
disp6.m
disp6.m~
e5~
eq_3_44.m
eq_3_44.m~
expt1.m
expt2.m
expt2_results
expt3.m
expt3_results
expt4.m
expt4.m~
expt5.m
expt5.m~
expt5_results
expt6.m
expt6.m~
expt6_results
expt6_results_old
expt6_results_old~
expt6p.m
expt6p.m~
expt_grad_confirm.m
expt_log_B.m
fcnchk.m
gendata.m
gendata.m~
generate_data.m~
get_approx_laplace.m
get_approx_laplace.m~
get_approx_laplace_old.m
get_probabilities_softmax.m
gp_data
gp_data_100
gp_data_20
gp_data_50
gp_function
gp_hyps
grad_3_44.m
grad_learn_hyps.m
gsl_test.c~
learnt_hyps
learnt_hyps_error
log_det_B.m
log_det_B.m~
mex_laplace_subroutine.c
mex_laplace_subroutine.c~
mex_laplace_subroutine.mex
mex_laplace_subroutine.o
mission.txt
octave-core
plot2.m
plot3.m
plot_conf.m
plot_f.m
points
points_100
points_20
points_50
sq_dist.c
test.m~
test1.m
test_prediction.m
timetable.txt
timetable.txt~
val_3_44.m
vis-selections.m
vis-selections.m~
vis_expt5.m
vis_expt5.m~
vis_selections.m
vis_selections.m~
vise6p.m
vise6p.m~
visualise.m
visualise.m~
writeup.latex

README

Large-scale Gaussian Process Classification, by Mark Norrish, supervised by Edwin Bonilla, 2012

USAGE
Octave or Matlab is required. Classification should work under either, but everything graphical probably only works with Octave. Open either of those and ype:

cross_validate()

to cross-validate a data set stored in a file called gp_data, whose hyperparameters are stored in a file called gp_hyps. These had better match the dimensions of the covariance function (func in cross_validate.m).
If you don't have hyperparameters already, you can try

cross_validate_with_learning()

which does much the same thing. If you want to accelerate learning, use only p points chosen by method m (see choose_subset.m for method indexing), type

CV_L_LDA(m, p)

. I recommend m = 0 at time of writing, 19/4/2012.
Currently, typing

expt6

will run the latest experiment I've been running.
You may notice that none of these actually works. This is because I've recently modified alg_3_3.m to try to do a bottleneck computation in parallel, and it's still in the debugging phase. To get the classifier working anyway, comment out the lines between the ones labelled "start comment" and "end comment".
Something went wrong with that request. Please try again.