Re: Corner of L-Curve


vivek wrote:
Greg Heath wrote:
<snip>

You might first want to find a way to detect the elbows in eta
and/or log10(eta) before smoothing rho and dealing with eta vs
rho.For example,

y = eta; dy = diff(eta); dy = [dy(1);dy];
d2y = diff(dy); d2y = [d2y(1);d2y];
Kappa = d2y./ (1+(dy).^2).^(3/2);

Plots of d2y and Kappa have maxima at the elbow.

Hope this helps.

Greg


Hi Greg,

Frankly speaking i'm a novice in the field of graphs and l-curves.
Now, I did run your bit of code which you gave as example and I did
see the maxima's for kappa and d2y. My question now is, do I need to
calculate rho and eta seperately

Calculate? They are already given.

or should I differentiate eta wrt
rho because i draw graph where x axis has rho and y axis has eta.

You gave us eta vs rho. It's up to you if you want that elbow or the
one that corresponds to eta vs index.

This is talked about by Jane in this particular topic in the 4th or
5th post.

How would you recommend me to find the solution of calculating the
corner of the L-curve ?

Smooth rho. Estimate Kappa from derivatives of eta vs rho. Look
for a local max.

Hope this helps.

Greg

.

Relevant Pages

  • Re: Corner of L-Curve
    ... as a function of rho (or rho as a function of eta, ... cubic spline at it, ... The bad news is I can't give you ...
    (comp.soft-sys.matlab)
  • Re: Corner of L-Curve
    ... I would like to know how can we calculate a corner of a L-curve. ... smooth rho. ... Separate plots of rho and eta show that eta and ...
    (comp.soft-sys.matlab)
  • Re: Corner of L-Curve
    ... If you plot eta ... as a function of rho (or rho as a function of eta, ... vivek wrote: ...
    (comp.soft-sys.matlab)
  • Re: Corner of L-Curve
    ... Greg Heath wrote: ... You might first want to find a way to detect the elbows in eta ... Plots of d2y and Kappa have maxima at the elbow. ...
    (comp.soft-sys.matlab)
  • Re: Subsets of cardinals in a well-ordering
    ... Let kappa be an infinite cardinal, ... Assume the result is true for regular cardinals. ... eta in A), xi in A), we have eta < xi and f< f. ...
    (sci.logic)