header image
«
»
[ # ] Unconstrained Optimization
August 13th, 2015 under Programming

MATLAB/GNU Octave offers simple ways to find the minimum of any function. The two easiest methods are

Click to toggle codeblock
1

fminsearch
fminsearch
and
Click to toggle codeblock
1

fminunc
fminunc
. The former uses a simplex search method while the latter uses a gradient based method.

Here is a simple example for a lognormal parameter estimation:

Click to toggle codeblock
1
2
3
4

x = sort([0.0027400 0.0093730 0.0458460 0.0135300])
y = (1:length(x))./(length(x)+1)
fun = @(p) sum( (logninv(y,p(1),p(2)) - x).^2)
estimate = fminsearch(fun,[-5;2])
x = sort([0.0027400 0.0093730 0.0458460 0.0135300])
y = (1:length(x))./(length(x)+1)
fun = @(p) sum( (logninv(y,p(1),p(2)) - x).^2)
estimate = fminsearch(fun,[-5;2])