Automatically detect and cast variables to expected type(s)

[GabeAl]

I have bumped into issues a few times where the expected formats for the inputs to biglasso turn out to be invalid, so I'd like to propose a set of usability enhancements that many could benefit from.

1. Matrix format. Suggestion: Detect normal matrix and cast to numeric big.matrix internally. (Example: I'm used to using normal matrices, and integer matrices. The former causes an error that is not clear, and the latter is a recipe for a crash! Immediate crash on small problem #22 )
2. If matrix is not big.matrix already, use the non-disk-spinning cast automatically or let user pick ( as.big.matrix() pegs spinny little disk for hours despite terabytes of RAM available kaneplusplus/bigmemory#97 )
3. Response format. Cast to numeric. (Example: I'm used to using factors for classification, but biglasso performs numeric operations on them even when doing classification. This leads to strange errors and does not let cv.biglasso complete correctly: Warnings/errors on run #25 ).

If changing the current behavior is not a good idea, it may still be nice to report an error about types so the user can make the necessary adjustments themselves.

Thanks for considering!

[privefl]

I don't know if @YaohuiZeng would have time to work on this as he started a new job.

Maybe you could make a PR.

[YaohuiZeng]

Sorry for late notice. And indeed I have limited time to maintain here. @privefl , I'd be more than happy to add you as a collaborator/contributor so you can directly pull changes if you'd like.

[YaohuiZeng]

Nonetheless, I'll try my best to find time to work on "turn off standardization", which is a frequent feature request. Other features, like here, will depend on my bandwidth, unfortunately.

[privefl]

Hi @YaohuiZeng,
Don't worry about me, I can just answer some basic questions about SSR and {bigmemory}.
I won't have much time to do anything else anyway.