[R-sig-Geo] Help regarding Autoencoder

[Roger Bivand]

On Tue, 24 Jan 2017, Subir Paul wrote:

> Dear All,
> I was using "SAENET" package for autoencoding.
> I used the following code:
> h=SAENET.train(data[,1:200], n.nodes = c(100, 20), unit.type =
> ("logistic"), lambda = 0.001,
>               beta = 0.0001, rho = 0.01, epsilon = 0.01, optim.method =
> c("BFGS"),
>               rel.tol=sqrt(.Machine$double.eps), max.iterations = 1000)
> But its showing the error:
> Error in optim(par = theta, fn = J.theta, gr = grad.theta.J, method =
> optimizer,  :
>  initial value in 'vmmin' is not finite
>
> Can anyone help me regarding this issue?

Try saying debug(SAENET.train) and stepping through to see what initial 
value is being passed to optim for vmmin. It may be an issue with your 
input data, or it may be something in the function that makes unmet 
assumptions about your data. Running traceback() after such an error 
probably will only show that optim is being passed a a non-finite value 
(divide by zero, log of zero?). Consider asking the package maintainer, 
and if possible provide a small, reproducible example.

Hope this helps,

Roger

>
> Thanks in advance.
>
> *With Regards*
> *Subir Paul*
> *Research Scholar (PhD)*
> *Water Resources & Env. Engg.*
>
> *Department of Civil Engineering*
> *Indian Institute of Science, Bengaluru*
>
> 	[[alternative HTML version deleted]]
>
> _______________________________________________
> R-sig-Geo mailing list
> R-sig-Geo at r-project.org
> https://stat.ethz.ch/mailman/listinfo/r-sig-geo
>

-- 
Roger Bivand
Department of Economics, Norwegian School of Economics,
Helleveien 30, N-5045 Bergen, Norway.
voice: +47 55 95 93 55; e-mail: Roger.Bivand at nhh.no
http://orcid.org/0000-0003-2392-6140
https://scholar.google.no/citations?user=AWeghB0AAAAJ&hl=en
http://depsy.org/person/434412