One more (better?) way to represent missing values

[shashi]

Just making sure this is on file in discussions here. http://wizardmac.tumblr.com/post/104019606584/whats-wrong-with-statistics-in-julia-a-reply "...the current NA and proposed Nullable approaches to missing values are a weak foundation that I would be hesitant to build upon."

The idea is you accompany the data vector with a frequency weight vector. A weight of 0 represents that the value is missing.

[nalimilan]

As noted in JuliaLang/julia#9363 (comment), that vector already exists as a BitArray. I guess some tests should be done to see whether the performance cost of BitArray over a plain array of boolean or integer is significant.

[johnmyleswhite]

I don't really think this package is the appropriate place to handle this problem, unless what we mean is that downstream functions should support frequency weights embedded as a column in a DataFrame (which I do think we should do).

[simonster]

DataArrays seems like the right place for this discussion. Let's discuss this further in JuliaStats/DataArrays.jl#133.