Re: good multivariate regression packages?
On 11/26/05, Dirk Eddelbuettel <edd@debian.org> wrote:
> | I should mention that this has led me to believe I should be looking
> | at nonparametric models, and have thus looked at using libsvm (not in
> | Debian), a Support Vector Machine library/tools. These seemed
>
> The e1071 package on CRAN has (some) SVM support. As I recall, there was also
> an R News article on it.
I recall looking at this, and as far as I can tell, these are either direct bindings to the libSVM package I already used, or simply offer an exact R-reimplementation of it. I assume the original/C implementation will be faster.
> | particularly speedy; 2) there is an outright overabundance of
> | regression methods in R, to the point where I am drowning in
> | information; with my meagre knowledge of regression methods I cannot
> | assess which methods would be most appropriate to my task, what the
> | tradeoffs, advantages and disadvantages are, etc. I made a similar
>
> Oh come, then R is even better as the methods are there, as are documentation
> and pointers to further reference.
>
> If you took your argument to its conclusion, you'd draw a straight regression
> line and be done with "because that's all we know and care about". Not.
:) I do suppose I've gotten a bit mentally lax and didn't do my due diligence. I've been spoiled with libSVM, which puts a strong emphasis on making itself very accessible and useful to the general user, one with no background knowledge in the underlying learning method. And this brings up a point: I think what I *really* was after in my original posting, although I didn't word it as such, is other packages in Debian that are capable of regression and have a similar low learning/usage curve. Learning the function I have illustrated is *not* on the critical path in my research, more of a side-project, hence I'm trying to limit the amount of time I invest there. This naturally makes me favour a "black box" regression method approach. Although I can see that a completely general regression method for 6D space might be quite involved and require some background knowledge, I was hoping that for a single specific case, such as the earlier illustrated function, might be achievable using a "black box" regression method.
> Lacking that, you could always peruse books like Hastie, Tibshirani,
> Friedman, "The Elements of Statistical Learning" but you probably already
> knew that :)
I don't actually know what the good Regression/Learning references are, although I presume the ones which have their own R packages (e.g., MASS) are worth a serious look. Any others that are a must? Are there any worthwhile online sources for getting a decent overview? I'm currently stuck at home to a fair extent (paternity leave), so getting at the references at the library might be a bit slow... I don't expect online sources to have anywhere near the depth of an academic monograph, but perhaps it might be useful to get my feet wet in the meantime...
--
Maciej Kalisiak
<mkalisiak@gmail.com>
http://www.dgp.toronto.edu/~mac
Reply to: