Consider making caret model list a named object

[zachmayer]

Consider making caret model list a named object

[zachmayer]

That's a good suggestion to make the fitting process more accessible. How would the tune grids for each method be specified? Maybe methodList could be a list of lists, where each element has either a tuneLength or tuneGrid parameter to pass to caret? Something like:
methodList <- list(rf=list(tuneGrid=data.frame(.mtry=c(2,4,8,1)), nnet=list(tuneLength=10)), knn=list(tuneLength=25))

I don't have a ton of time to work on this at the moment, but will add it to my todo list. In the meantime, I'd be happy to merge in any pull requests you have the time to write.

[jknowles]

I'll probably hit you up with a pull request if you think this is a good idea. It's poor form to suggest an enhancement and at least not give a pull request a stab. I think you've built a great structure here. The caret API has changed a bit since you worked this up, so I forked the repo and have been working on making sure it works with caret 6.0+ models.

You are absolutely right about the tune grids. Making it a list of lists is a really good fix for that.

[zachmayer]

I think it's a good idea. This package hasn't seen a lot of love from me lately, and definitely needs some updating to work with caret 6.0.

[jknowles]

I think the next step is to generate some print and summary methods for both caretEnsemble objects and caretList objects. I have some ideas, but am happy to incorporate any thoughts you have about what we should show the user.

[zachmayer]

I don't currently have any ideas for print and summary, so feel free to
code up your ideas. If that sparks anything on my end, I'll comment on
your PR.

On Fri, Feb 28, 2014 at 11:24 AM, Jared Knowles [email protected]:

I think the next step is to generate some print and summary methods for
both caretEnsemble objects and caretList objects. I have some ideas, but
am happy to incorporate any thoughts you have about what we should show the
user.

Reply to this email directly or view it on GitHubhttps://github.com//issues/2#issuecomment-36367426
.

[jknowles]

@zachmayer I think you can close this issue. If we want to revisit the caretList objects and further enhance them, that should be a new issue.

[zachmayer]

A note for future reference: we've renamed buildModels to caretList, and it's become a core part of the package. Everyone who's using caretEnsemble seems to be writing giants loops to fit "every model in the caret package," so it seems like a good idea to have a unified way to fit all those models and make sure they'll work with caretEnsemble, without issues from re-sampling indexes or random seeds.

(maybe we can write a vignette on why this is a bad idea and a few hand-tuned models is usually better...)