Enhancements for *dply

[krlmlr]

• Generate .id column as factor in ldply(); fixes ldply and the like should create the .id column as factor, preserving the order of the names in the list #137
• New parameter .id for rdply() to specify name of index column (default: .n); fixes Allow renaming of .n column generated by rdply #142
• New parameter .id for ldply() to specify name of index column (default: .id); fixes Allow renaming of .id column generated by ldply #140; fixes plyr does not preserve names in nested calls #107

Rebased and edited #177.

[hadley]

Thanks! What do you do think about something similar for adply where it would form the prefix for the column names? Not sure what happens currently.

[hadley]

Unfortunately changing .id to a factor breaks ggplot2.

[krlmlr]

Could you please provide more detail? Which part of ggplot is broken?

I came up with this issue because I frequently have the following situation:

1. A sequence of values given in a vector (or a list)
2. This sequence results in a data frame after a call to ldply
3. This data frame is used in a ggplot where the .id column is mapped to the facet label

Previously, this resulted in a sorting of the facets, but mostly I want to specify the sequence of facets once (in step 1). As a workaround, I have invented a function ofactor that works like factor but orders the levels in order of appearance (and not in sorted order as with factor). Using a factor for the .id column resolves this issue in an elegant way.

[hadley]

Facetting, because it expects a character, and it's now a factor. I think it's a reasonable change, but plyr is used by so many packages that even changes that make sense may cause bugs in existing, and that's something I'd prefer to avoid if possible.

[krlmlr]

I have searched ggplot for .id, and didn't find anything. Do you mean user code that calls ggplot with the result of a ldply call? Could you give a failing example? I used to think that factors were always coercible to character...

[krlmlr]

A simple test reveals the error: http://rpubs.com/krlmlr/11768

Of course we could patch ggplot2, but perhaps the safest way is to ensure that a factor is returned only if the user requests it, say, by passing a value to the .id parameter. New default for the .id parameter could be NA to indicate that the old behavior is requested. ?

[hadley]

That seems reasonable to me. Would you mind submitting a patch?

[krlmlr]

Done: #195