R help - Oct 2005 - ipop (kernlab) gives pars < lower bound ?

hi everyone,

ipop very quickly and accurately identifies the correct parameters in
a toy dataset i built, but when i use ipop on the real dataset i get
values for the parameters  " primal(res) "  that are less than zero,
even though i specify zero for the lower bound :  l = rep(0,
length(c)) , where length(c) is the number of parameters i'm trying to
identify.

the parameters are not A LOT less than zero-- they're close to zero,
but still too large a problem to ignore. (eg., i get a value X1 -0.006 -- it
wouldn't be a problem if X1 was btw -1 e-5 and zero, but
it's not.)
can anyone suggest a remedy?

i'm wondering if this could be caused by where i've set my
"range" and
my "bound".  i've played a little with range, but it hasn't
helped.  i
don't know how to use "bound"-- i don't understand the info on
"bound"
on the ipop Rhelp page.
"sigf" is at default 7 sig figs, and the negative numbers are not
within that threshold of zero.

i'm also wondering if this problem could be related to the warning
messages i receive when i run ipop :

Warning:
number of columns of result
       not a multiple of vector length (arg 2) in: rbind(rho, tau, alpha, nu)

i get this warning message in both the toy dataset and the real
dataset, though as i mentioned i only get negative (illegal) parameter
values with the real dataset.
i don't know why i get this message, or what to do about it.

any advice is very much appreciated!

alexis

hi again,

i have fixed some of the problems i emailed about previously-- the
answer is to fiddle with the "margin" variable.  but new problems have
arisen: when i set the "margin" variable appropriately to preclude
negative values, i get the following error message:

Error in drop(.Call("La_dgesv", a, as matrix(b), tol, PACKAGE =
"base")) :
system is comptuationally singular: reciprocal condition number = 1.57163 e-16

and i was wondering if there's anything i can do about it, or if my
data is just not good for this problem.

also, i am still curious about
(1) how to use ipop's "bound" argument

(2) what should i make of the warning message discussed below?

thanks again,

alexis

On 10/18/05, Alexis Diamond <alexisjdiamond at gmail.com>
wrote:
> hi everyone,
>
> ipop very quickly and accurately identifies the correct parameters in
> a toy dataset i built, but when i use ipop on the real dataset i get
> values for the parameters  " primal(res) "  that are less than
zero,
> even though i specify zero for the lower bound :  l = rep(0,
> length(c)) , where length(c) is the number of parameters i'm trying to
> identify.
>
> the parameters are not A LOT less than zero-- they're close to zero,
> but still too large a problem to ignore. (eg., i get a value X1 > -0.006
-- it wouldn't be a problem if X1 was btw -1 e-5 and zero, but
> it's not.)
> can anyone suggest a remedy?
>
> i'm wondering if this could be caused by where i've set my
"range" and
> my "bound".  i've played a little with range, but it
hasn't helped.  i
> don't know how to use "bound"-- i don't understand the
info on "bound"
> on the ipop Rhelp page.
> "sigf" is at default 7 sig figs, and the negative numbers are not
> within that threshold of zero.
>
> i'm also wondering if this problem could be related to the warning
> messages i receive when i run ipop :
>
> Warning:
> number of columns of result
>        not a multiple of vector length (arg 2) in: rbind(rho, tau, alpha,
nu)
>
> i get this warning message in both the toy dataset and the real
> dataset, though as i mentioned i only get negative (illegal) parameter
> values with the real dataset.
> i don't know why i get this message, or what to do about it.
>
> any advice is very much appreciated!
>
> alexis
>