annotate general/numerical/scalar/hypdecay.m @ 53:3ba80c9914ff
Minor doc fix, added .class to .hgignore
| author |
samer |
| date |
Mon, 02 Feb 2015 10:47:55 +0000 |
| parents |
db7f4afd27c5 |
| children |
|
| rev |
line source |
|
samer@4
|
1 function ret=hypdecay(k,d)
|
|
samer@4
|
2 % hypdecay - compute interpolation ratio for hyperbolic decay
|
|
samer@4
|
3 %
|
|
samer@4
|
4 % hypdecay :: [Size->nonneg], [Size->nonneg] -> [Size->nonneg].
|
|
samer@4
|
5 % hypdecay :: [Size->nonneg] -> ([Size->nonneg] -> [Size->nonneg]).
|
|
samer@4
|
6 %
|
|
samer@4
|
7 % this gives 'hyperbolic' convergence which is more like
|
|
samer@4
|
8 % a sort of diffusion by Brownian motion. The trick is
|
|
samer@4
|
9 % to add a constant to the inverse of each natural parameter,
|
|
samer@4
|
10 % which is like a temperature or variance. The constant
|
|
samer@4
|
11 % is like a diffusion constant
|
|
samer@4
|
12 %
|
|
samer@4
|
13 % This function supports partial application: if only one
|
|
samer@4
|
14 % argument is supplied, it returns a function handle.
|
|
samer@4
|
15
|
|
samer@4
|
16 if nargin==2, ret=1./(1+k*d);
|
|
samer@4
|
17 else ret=@(d)1./(1+k*d); end
|
