salbert83

Also, regarding this line: "posterior = sample(model, NUTS(),3000);" perhaps a comment that this could take some time depending on one's machine. Thanks for making this preview available.

One more, replace all_words_text = StringDocument(string([string(word, " ") for word in all_words]...)) with all_words_text = StringDocument(join(all_words, " ")) Thanks

Also, think you're missing the line to load the file, something like raw_df = DataFrame(CSV.File(email_path)) Thanks

The short answer is no, because the solution to the associated optimization problem might not be unique. I extended the writeup. [QuantileCalc.pdf](https://github.com/JuliaStats/StatsBase.jl/files/6381279/QuantileCalc.pdf) I began looking at this in relation to...

1) Please review the PDF, where I examined the 30% unweighted quantile vs my implementation on the integers 1..100. 2 & 3) I believe my implementation satisfies these requirements, unless...

The implementation largely follows the current implementation. A view is taken restricted to observations with nonzero weights, so (3) should be satisfied in all cases. The code should also be...

The unweighted quantile is the one from Statistics.jl, right? I see Julia defaults to definition 7 of Hyndman and Fan (1996). It seems definition 4 is more consistent with the...

Statistics.quantile already supports the optional arguments, julia> z = [1:100...]; julia> quantile(z, 0.3, alpha = 1, beta = 1) # These are the defaults if left unspecified 30.7 julia> quantile(z,...

I implemented the MLE consistent calculation using an optional kwarg. It doesn't cover the full range regarding alpha and beta settings available in Statistics.quantile, but does provide an option for...

While I think alpha = 0 and beta = 1 are consistent with the optimization approach, I think it would be misleading. Statistics.quantile supports arbitrary alpha and beta in [0,1]...