[GSoC] Dataframes for D

[Prateek Nayak]

On Tuesday, 25 June 2019 at 17:54:36 UTC, jmh530 wrote:
>
> Glad to see you're still making great progress.
>
> I had worked on the byDim function in mir.ndslice.topology is 
> byDim because I had wanted the same sort of functionality as 
> R's apply. It works a little differently than R's, but I find 
> it very good for a lot of things. Your version of apply (I'm 
> looking at the apply branch of magpie) looks like it operates a 
> bit like a byDim chained with an each, so byDim!N.each!f. 
> However, it also has this index variable allowing it to skip 
> rows or something (I'm not really sure if this feature pulls 
> its weight...).
>
> So I have two questions: 1) does byDim also work with 
> dataframes?, 2) can you add an overload that is apply(f, axis) 
> without the index parameter?
>
> One of my take-a-ways from looking at the apply function (again 
> just looking at that apply branch) is that you might benefit 
> from using more of what is Ilya has already put in mir.ndslice 
> where available. For instance, the overload of apply that is 
> just apply!f is basically the same as mir's each, but each has 
> more features.

1) Current, byDim doesn't work on DataFrame DataFrame.
2) Sure, the overload can be made but what are you specifically 
looking for?
apply(f, axis)(indexes) ?

You are right, apply works like byDim!axis.each on particular 
columns/rows.
I'll look into Mir's implementation. Thanks for that advice. I do 
believe apply can be strengthened to account for different use 
cases.
When the heterogeneous DataFrame support came, mir-algorithms was 
dropped from dependencies and Structure of Array implementation 
was taken up using TypeTuples. Once the basic working is solid, 
I'll port useful features from Mir to Magpie.