TimeSeries.jl
TimeSeries.jl copied to clipboard
`Base.view` support
@view ta[1:42]
to avoid allocation
Maybe we need a new type for viewed TimeArray.
Hi, is there an ETA on this functionality? We have integrated TimeSeries in our package, but when slicing data, there is too much memory allocation with respect to DataFrames. We would like to keep TimeArrays, but the performance hit is significant.
Hi @jd-lara , no ETA at this moment. I can mark this issue as high priority in my mind. I still need to do some surveys about the implementation details of @view.
Thanks, we will stick with TimeArrays for now and hopefully won't be too much of a bottleneck.
I wrote some study notes here:
- [ ] Design doc of Julia's native
SubArrayis here: https://docs.julialang.org/en/v1/devdocs/subarrays/index.html - [ ] We might need a new data type
SubTimeArray, in order to provide index translation on the top ofgetindexsyntax.- e.g.
SubTimeArray[1:10, :col]won't copy.
- e.g.