TimeSeries.jl
TimeSeries.jl copied to clipboard
JuliaStats
Use Nullable Array package
This has just gotten registered so just want to throw this out for possible implementation down the road. https://github.com/JuliaStats/NullableArrays.jl
Anyone working on this right now? I need left/right-joins and it would make sense to implement this first - I can do it this weekend if no one else has started on it.
On second thought, this may have farther-reaching consequences than I thought, so I might just implement what I need with NaNs for now... I'll push my work to a new branch for reference, although I know the convention so far has been to avoid intentionally introducing NaNs, so we may not want to merge it until we can do it properly with Nullables.
Yeah, I was hoping with Julia 0.4.0 and Nullable it would be simple and easy from there. Turns out there is a performance hit. I'm in favor of using NaN for now, and making this part of version 0.7.0 changes.
The way to not disrupt the API too much by introducing NaN sentinels is to make it a kwarg I think, with the default being no padding with sentinels, since that's been the behavior so far.
So, for example, lead/lag would gain padding=false keyword args? I like that.
I worked on developing the NullableArrays package. I'm happy to help navigate performance issues, if I can. If there are requests for features, please do file an issue.
Thanks @davidagold. Can you update us on how the package development is coming along. We're about to do some other important changes that but the implementation of missing/unknown/consumed values is coming up next.
The package currently is more or less at feature parity with DataArrays, modulo the PooledDataArray interface/functionality of the latter. That functionality is to be refactored into a different package that will build support for factors on top of NullableArrays.
You should probably support both standard Array and NullableArray, as the latter will always have a performance hit. When people don't need missing values, they would use the former.
Do you mean enhance the existing TimeArray type's value element or a new type?