add filtercols and filtercols!

[bkamins]

We have mapcols so we can consider adding filtercols (with view kwarg) and filtercols! that would work like filter but would filter columns. There is no rush to add this.

[zerefwayne]

@nalimilan @bkamins This looks interesting. Can I work on this?

[bkamins]

sure - just make sure to understand the API of filter and filter! first. Also this should be synced with #2417 as the API should be the same in both places.
The question is if the predicate should get a column or a "column name", column Pair. I think it should be just a column

[bkamins]

ah - and filter should support copycols kwarg additionally.

[bkamins]

And the last thing - maybe wait for 0.22 release before opening a PR. Then you will be sure that you are working on a clean master (of course you can start implementation now if you want; 0.22 should be tagged this week). Finally - have a look at https://github.com/JuliaData/DataFrames.jl/blob/master/CONTRIBUTING.md, as following this will speed up the review process.

[zerefwayne]

I'll have a look at filter first. It will roughly take me a week to implement this, I'll open the PR after syncing with 0.22! Thank you for the heads-up @bkamins!

[bottine]

I don't know if that's reasonable wrt orthogonality and API, but I would have liked to have mapcols!(fun, df, cols) where cols is the usual column selector, allowing to apply a function only on specific columns. Could e.g. have mapcols!(uppercase, df, names(df, String)) to make string values uppercase.

[bkamins]

Now the pattern for this is:

transform!(df, names(df, String) .=> ByRow(uppercase), renamecols=false)

I understand you find mapcols! shorter or easier to reason about?

Note that you would need to write:

mapcols!(col -> uppercase.(col), df, names(df, String))

though.

[bottine]

Ah, indeed, I didn't think it would be so short with transform!, you're right!
But then,

mapcols!(fun, df) == transform!(df, All() .=> fun, renamecols=false)

so the point of mapcols! is mostly that it's a bit shorter?
(Btw, let this be an opportunity to thank you for the package!)

[bkamins]

Yes, the original point is that mapcols was introduced much earlier in the development process of DataFrames.jl than transform.