quantile across dimension

[dvd101x]

Hi, while working on #1552 I think this is the minimum changes to make it work, but maybe there are some topics to discuss.

1. By researching quantileSeq, I've seen some other references and I think it could be changed to just quantile
2. A different method could be implemented for example linear (numpy's default)
3. It could be refactored to use more of typed capabilities
4. utils/lastDimToZeroBased doesn't apply directly for this function, in my opinion is too specific.
5. Other stat functions are also missing dimensions functionality: mad, median, mode and prod
6. It seems like other stat functions could benefit from apply to reduce the amount of code

Please let me know if any of these topics could be addresses in this PR.

[josdejong]

The PR looks good and well tested, thanks David. I think we can merge it as-is.

About your discussion points:

1. Renaming quantileSeq to quantile: I don't remember why the function was originally named like this. I would prefer not "just" renaming the function, that is a breaking change again. But if there is a good reason for it we can consider it. Is the name quantile more common in other applications for example?
2. I'm not sure what you mean with A different method could be implemented for example linear (numpy's default)
3. Would you like to change lastDimToZeroBased to make it more usable in different cases? Do you have a proposal?
4. The function quantileSeq can indeed use some refactoring if you want to give that a go. A PR would be welcome, but let's please create a separate PR for this.
5. Yes I would love to add the dimension argument to other functions too. (let's do that in a separate PR :))
6. Definitely! You're full of ideas to improve the code base :). A separate PR with a refactor would be nice.

[dvd101x]

Thank you, Jos,

Ok, that sounds good to have those topics on a different PRs. I think I left some unwanted modifications in the docs which I will fix soon.

1. After further review, I noticed quantileSeq has a parameter that is not common for other quantile, If probOrN <= 1 then it is prob as other quantile but if probOrN is an integer > 1 then it acts differently by splitting the range from 0 to 1 in N places and returning an array of prob and the rest works like other quantile. Here are some references of other quantile that are similar between them:

• https://www.rdocumentation.org/packages/stats/versions/3.6.2/topics/quantile
• https://numpy.org/doc/stable/reference/generated/numpy.quantile.html
• https://octave.sourceforge.io/octave/function/quantile.html
• https://www.mathworks.com/help/matlab/ref/quantile.html
• https://reference.wolfram.com/language/ref/Quantile.html
• https://docs.julialang.org/en/v1/stdlib/Statistics/#Statistics.quantile!

2. In numpy there are about 9 different methods to compute quantile. The linear method instead of giving the average of two values when it falls in between, does a linear interpolation.
3. Currently lastDimToZeroBased is limited to where the number of arguments is exactly two. quantileSeq has more than two arguments so it can't use it as such.
4. Ok, I will try
5. Ok, thanks!
6. Ok, thanks! :D

[josdejong]

Thanks for the reply and all your inputs of this week and last week on the project David, I'll look into them after the holidays, sorry for the delay.

[dvd101x]

Hi Jos, thanks for the reply

Please don't worry about it, and have great holidays!

[josdejong]

Published now in v11.10.0