Attributes:
dist=[]; -- type of distance (SEE BELOW)
child=kernel; -- if nonempty, generate distance from this kernel
rather than from type described in dist
distparam=[] -- parameters of the distance
kercaching=0 -- yes if caching the distance (only for data_global)
calc_on_output=0 -- calc distance on outputs (Ys), rather than inputs (Xs)
output_kernel=0 -- output associated kernel (if any) rather than distance
dat=[] -- storage of data (for use when training/testing dists)
Methods:
calc(d,d1,[d2]): calc inner products in feature space between data d1
and d2 (or itself if d2 not specified) using distance d
train,test
DISTANCE PARAMETERS & DESCRIPTION
euclid d(x,y)=||x-y||_2 (same as using child=kernel)
norm d(x,y)=||x-y||_p, p can be inf, -inf see help norm
template_distance example to help make your own distance
custom takes values from indices of matrix (distparam)
custom_fast takes values from global variable K
Example: calc(distance,data(rand(5)))