Finixh #58 by adding the vardim keyword argument to Tables.matrix, al…

[quinnj]

…lowing the user to specify whether input columns should be materialized as matrix columns or rows

[nalimilan]

Thanks, looks good! My only hesitation is whether we should call the keyword argument vardim. The only place where that term is used currently AFAIK is in StatsBase's weighted cov, but it's going to be changed to dims for consistency with Statistics.cov. That said, vardim sounds clearer (dims would look weird here) and it could be used for JuliaStats/Clustering.jl#79.

[andreasnoack]

I think dims and vardim are equally confusing but since the reductions in Base use dims, we have been switching from vardim to dims in StatsBase (although the switch is not yet complete) so I think it would better to use dims because of the precedence.

[quinnj]

Updated from vardim to dims.

[nalimilan]

That naming issue is quite annoying. After thinking more about it, I feel like using dims=1 to mean "transform columns to rows" is quite arbitrary. We could as well say dims=2 means "transform rows to columns", or any other intepretation. There's actually no clear relationship with reductions which use the dims argument in Base.

Maybe transpose::Bool=false would be clearer? CSV.file already supports that argument.

[quinnj]

So what should we do here? I don't have strong opinions.

[nalimilan]

I think I prefer transpose.

[tlienart]

I think transpose makes sense here. at the end of the day if this is well documented, it should be fine.