[no subject]

[ Thread Index | Date Index | More lists.tuxfamily.org/eigen Archives ]

Best regards

Content-Type: text/html; charset=ISO-8859-1
Content-Transfer-Encoding: quoted-printable

Hello Thomas<br><br>I am really enjoying Non Linear Optimization module. It=
 lets me solve very complex things in a smart way :-) .<br><br>However I ha=
ve had a hard time this weekend about a 17 unknowns set of equations minimi=
zed with Levenberg Marquardt.<br>
<br>With initial values very close to results it worked well, but putting r=
ealistic initial values the system did not converge. It returned a TooManyF=
unctionEvaluation (5),<br><br>Checking the code, it seems that in the const=
ructor, the max number of function evaluations (fev) is:<br>
<br><span style=3D"font-family: courier new,monospace;">=A0=A0=A0 </span><s=
pan style=3D"font-family: courier new,monospace;">maxfev (400)</span><br><b=
r>But old lmder1() function, before calling minimize(), inits some paramete=
<br style=3D"font-family: courier new,monospace;"><span style=3D"font-famil=
y: courier new,monospace;">=A0=A0=A0 n =3D x.size();</span><br style=3D"fon=
t-family: courier new,monospace;"><span style=3D"font-family: courier new,m=
onospace;">...</span><br style=3D"font-family: courier new,monospace;">
<span style=3D"font-family: courier new,monospace;">=A0=A0=A0 parameters.ma=
xfev =3D 100*(n+1);</span><br><br>So in my case, maxfev =3D=3D 400 instead =
of maxfev =3D=3D 100*(17+1) =3D=3D 1800.<br><br>Doing this change, the syst=
em converged with good results after nfev =3D=3D 1619..., What a relief !<b=
<br>In <a href=3D"http://eigen.tuxfamily.org/dox-devel/unsupported/group__N=
upported/group__NonLinearOptimization__Module.html</a> you state that all m=
inpack methods like lmder1() are deprecated.<br>
<br>From your experience, could you include the better way to initialize th=
e params., adapted to the number of equations and unknowns?<br><br>Best reg=


Mail converted by MHonArc 2.6.19+ http://listengine.tuxfamily.org/