Simon Kornblith

The right way to do this is to implement `datahcat`, `datavcat`, and `datahvcat` and make the `data` and `pdata` macros call these. I have this working, but 1) if we...

The bug is that setting a portion of a DataArray to be another DataArray fails if the latter has NAs. #62 has an easy fix that would unfortunately create a...

(And after that change, my tests are still failing, now because of JuliaLang/julia#5448.)

Unfortunately, as my benchmarks show, it's possible to extract reasonable performance out of the BitArray in this benchmark, but abstracting that would basically require using a macro for looping so...

We can use a different bit pattern for NULL NaN, as @tshort commented in DataArrays.jl#5. I think 25% slower is about the cutoff where there'd be reason to have DataArrays...

There are 22 bits of payload for Float32 NaNs and 51 bits of payload for Float64 NaNs, and I don't think the CPU ever sets them to anything but zero...

@tshort Without any tweaks I think that would be a `Vector{Any}` holding 3 Ints and a Nullable. If we want it to be a `Vector{Nullable{Int}}`, then we'd need to define...

We don't currently implement `max` in `operators.jl`, so we are falling back on the implementation in Base, which evidently changed between 0.2 and 0.3. We should implement `max` in `operators.jl`.

I think the problem @johnmyleswhite is referring to is that, if you run `round(@pda [1.01, 1.02])`, the pool of the resulting PDA should be `[1.]` and not `[1., 1.]`. In...

I'm not sure, although it seems safer to use a DataArray even when working with floats. With a DataArray you could potentially have values that are `NaN` but not `NA`,...