The official maintainers were dismissive when I suggested there were some
problems I could fix with the then implementation of polr. I haven't looked
at it since, sorry.
On Tue, Aug 12, 2014 at 7:44 PM, Guido Biele [via R] <
ml-node+s789695n4695392...@n4.nabble.com> wrote:
> I modified (where ne
This could well be out of date because I have not paid any attention to the
official POLR code in a year or more. Attached is fixed-polr.R.
cheers,
Tim
On Wed, Oct 24, 2012 at 10:38 PM, ahs [via R] <
ml-node+s789695n4647311...@n4.nabble.com> wrote:
> Great!
> You can skip my question about s0 th
Great!
You can skip my question about s0 though, I found where it is being used,
but I still struggle with the code and convergence problem.
I also found something here
http://biostat.mc.vanderbilt.edu/wiki/pub/Main/CharlesDupontStuff/newPolr.R
that seems like someone tried to fix it, but with th
Sorry, I have been away. I'll have a look at this and get back to you
tomorrow.
Tim
On Thu, Oct 18, 2012 at 10:35 PM, ahs [via R] <
ml-node+s789695n4646600...@n4.nabble.com> wrote:
> Hello,
> I am trying to use this fix for the convergence problem in polr, but I
> don't seem to get the change of
Hello,
I am trying to use this fix for the convergence problem in polr, but I don't
seem to get the change of code right. I redefine the function polr by the
lines
tjb wrote
> if(missing(start)) {
> # try something that should always work -tjb
> u <- as.integer(table(y))
> u
Note that that the enhancements in my original post solve the unresolved
problem of Chaehyung Ahn (22 Mar 2005) whose data I reproduce:
y,x,lx
0,3.2e-02,-1.49485
0,3.2e-02,-1.49485
0,1.0e-01,-1.0
0,1.0e-01,-1.0
0,3.2e-01,-0.49485
0,3.2e-01,-0.49485
1,1.0e+00,0.0
0,1.0e+00,0.0
1,3.
The start value generation code in polr is also known to fail quite
frequently. For example, against the Iris data as recently posted to this
list by blackscorpio ( Sep 6, 2010).
> polr(Species~Sepal_Length+Sepal_Width+Petal_Length+Petal_Width,data=iris)
Error in polr(Species ~ Sepal_Length + Se
In polr.R the (several) functions gmin and fmin contain the code
> theta <- beta[pc + 1L:q]
> gamm <- c(-100, cumsum(c(theta[1L], exp(theta[-1L]))), 100)
That's bad. There's no reason to suppose beta[pc+1L] is larger than
-100 or that the cumulative sum is smaller than 100. For pr
8 matches
Mail list logo
