[R] Modelos neural netwok

Richard O'Keefe r@oknz @end|ng |rom gm@||@com
Sun Jul 28 00:18:36 CEST 2019


Splitting a data set into a part to be used for fitting a model
(the training set) and a part to be used for evaluating the
quality of the model on new cases (the test set) has been good
practice for a long time.  If the architecture of the model is
to be learned as well as parameters, a three-way split is used.
https://en.wikipedia.org/wiki/Training,_validation,_and_test_sets
will get you started.

I have not seen the book or the package.  With any example in
any book, you have to ask "what is this example supposed to
demonstrate?"  A bit over 40 years ago a friend of mine had
great trouble trying to learn statistics.  He read the books
diligently, but said to me "I look at the examples, and can
never understand why the method shown in the example is the
best method for that problem".  That's because it wasn't.
The example is meant to demonstrate a particular method, and
a problem is chosen that the method can be applied to, and
the author never bothers with the question "is this a good
method for this problem".  It may be that you are looking at
an example expecting to be told "what should I do with a
data set like this" whereas the author was telling you "this
is how to fit a model like this and use it to predict the
future".

This applies to examples in R documentation too.


On Sun, 28 Jul 2019 at 05:42, Agustín Alonso Rodriguez <
aalonso using rcumariacristina.com> wrote:

> Estimados miembros de la Lista:
>
>
>
> Me dirijo a vosotros para que, si podéis, iluminéis mi confusión.
>
>
>
> Estoy trabajando con los modelos neural netwok del  libro Forecasting,
> Principles and Practice, de Hyndman y Athanasopoulos, segunda edición.
>
>
>
> Mi confusión se debe al hecho de que para trabajar con estos modelos, se
> recomienda particionar la serie temporal en estudio en, al menos, dos
> submuestras:  una para estimar el modelo, y la otra para validar el modelo.
> Pues bien, en los ejemplos del mencionado libro y en el paquete fpp2, que
> acompaña al libro, los modelos se obtienen utilizando la muestra completa y
> haciendo predicción hacia el futuro. No veo ninguna alusión  a la  muestra
> de validación, antes aludida.
>
>
>
> Pienso que algo se me está escapando, y agradeceré cuanto podáis hacer para
> ayudarme.
>
>
>
> Muchas gracias por vuestra atención.
>
>
>
> Agustín Alonso
>
>
>
>
>
>
>
>
>
>
> ---
> Este correo electrónico ha sido comprobado en busca de virus por AVG.
> http://www.avg.com
>
>         [[alternative HTML version deleted]]
>
> ______________________________________________
> R-help using r-project.org mailing list -- To UNSUBSCRIBE and more, see
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide
> http://www.R-project.org/posting-guide.html
> and provide commented, minimal, self-contained, reproducible code.
>

	[[alternative HTML version deleted]]



More information about the R-help mailing list