Allow replacing with results of a function

[paulvictor]

Create another type which can replace with results of evaluating a function, optionally providing the data array itself as an argument. Use case: Replace with samples drawn from the distribution of non NA values.

[nalimilan]

Thanks, but I'm not sure a dedicated function and type are really needed here. I find an explicit loop clearer, and almost as short as your version:

[isna(dv, i) ? mean(dropna(dv)) : dv[i] for i in eachindex(dv)]

or the in-place version

for i in eachindex(dv)
    if isna(dv, i) dv[i] = m end
end
``

[paulvictor]

True, but that's the case with each_replacena too, here the example given seems simple, but given that a function can be used opens up a number of use cases, like replacing NA's with values drawn from a distribution described by the non-NA values(which need not be the same for each invocation), here again, this can be replaced by the transformation as you described, but I find this style better, since you make it more declarative(by passing the function(the what)) than imperative(generating and assigning the values(the how)). On a note aside, there are tests failing on travis and the coverage has been reported to be down, but no test errors occur when I build locally, is there anything I'm missing? Attaching a screenshot of how the tests ran, please give your feedback on the same.

[nalimilan]

Others should tell whether they find it useful before you do more work on it, but I think you should at least merge each_replacenawithfunctionresult into each_replacena. You could simply add a method for Callable arguments.

I'm not sure why the tests are failing at the moment.

[paulvictor]

@nalimilan , Can you check https://github.com/paulvictor/DataArrays.jl/commit/365f1a34235bd79e8b085170082e516abf328cdf, if that's what you meant? I can provide a squashed commit later if the code makes sense.

[nalimilan]

Yeah, something like that, but as I said I'll let other comment.

[paulvictor]

Thanks @nalimilan , will wait too...