DataFrame pivot/melt-cast/etc.

[HarlanH]

Functionality to take a DF in wide form and make it long, and vice-versa

[HarlanH]

we've got stack/unstack. Someone should sit down and try to figure out if that's good enough or if we need additional functionality.

[StefanKarpinski]

I'm sure you guys are aware of the [reshape] and [reshape2]. Also worth considering is the find/sparse functions in Matlab that kind of do similar things and are a really classic design.

[johnmyleswhite]

I find that the stack / unstack API is much less helpful than the cast / melt API. At some point, I'd like to go through and clean this up.

[tshort]

The existing stack(df, ["colX", "colY"] is equivalent of Hadley's melt(df, measure.vars = c("colX", "colY"). I added versions of melt and melt_df that use stack and stack_df. The only difference really is that the melt functions prefer id_vars.

That was the easy part. cast is the tough one.

[johnmyleswhite]

It would be really nice to use id_vars instead. That's just a set complement, right?

[tshort]

Yes it is. That's what the new melt does. If df has columns x1 through x4, melt(df, ["x1", "x2"]) is the same as stack(df, ["x3","x4"]). Both functions also allow three arguments.

[johnmyleswhite]

Ah. Much better! This is looking really promising.

[tshort]

More changes here. I added a simplistic pivot_table function. The purpose is like Hadley's dcast or Wes's pivot_table. The format is pivot_table(d, R, C, D, fun) where R and C are vectors indicating columns that are to be pivoted to either rows or columns in the results. D is the column to take for data. fun is the function to apply to aggregate (defaults to mean if left off).

[tshort]

Also, I changed to argument order for unstack to match that of pivot_table.

[tshort]

I'm closing as we've got most basics between stack, unstack, melt, and pivot_table. pivot_table is like Hadley's cast. It's still fairly limited, so folks can add specific feature requests.