| |
RSC_MDS reduced set construction - using MDS
|
a = rsc_mds(alg,hyper)
generates a rsc object, using the MDS pre-images method
hyperparameters:
child=svm algorithm worked on
rsv=.5 if rsv e [0,1): #rsv = #sv * rsv
if rsv > 1: #rsv
nn=10 neighbours to use for distance constraints
map='rbf_map' supply function name to be called for calculating
input space distances from feature space distances
must be supplied in rsc_mds directory
model:
alpha new alphas for rs-vectors
Xsv rs vectors
stats:
w2=0 final value of ||w-w*||^2
delta_w=0 total decrease in ||w-w*||^2
dw=0 run through delta_w
methods:
train constructs a reduced set, returns trained rs-machine
test tests new rs-machine on supplied data
example:
d=gen(toy2d('2circles','l=100'));
[r,a]=train(svm({kernel('rbf',1),'C=10000','alpha_cutoff=1e-2'}),d);
[r,a2]=train(rsc_mds(a,'rsv=.9'),d);
test(a2,d,loss)
remark: supports only rbf kernels up per default!
author: Goekhan BakIr, Philipp Berens
reference: The Pre-Image Problem in Kernel Methods, J.T.Kwok, 2004