[Rd] Withdrawal of package Rigroup-0.84.0 from CRAN

[Kevin Hendricks]

Hi Brian and Uwe,

Given my package passes all checks but I have received no further responses since your initial reply, I assume you are no longer interested in the package Rigroup-0.84.0.  Your concerns about my chosen license being too vague "GPL | LGPL" is in direct contradiction to what your own documents say about not changing your license after a package has been accepted at CRAN.  This is the same license this package has had since it was first accepted in 2006 ("GPL or LGPL your choice").  Your additional requirement was that I explain why I did not reply to your January e-mail thereby forcing you to do "extra work to archive the package" was already met on this very list. As I explained back in March, I never received it (and no Uwe that does not mean I am claiming you never sent it - just that I did not receive it!).  Not quite the "thank you for contributing to R" I hoped for, but after 7 years or so, it is unfortunately what I expected... 

Therefore this is to let you know that I am hereby withdrawing my package Rigroup-0.84.0 from CRAN consideration.  You can obviously keep using the previous (now-archived) version under its license.  But since I will no longer be supporting the package and as its author I ask you to remove all versions from CRAN.  This is of course your choice given the original license Rigroup was released under. If anyone's package depends on Rigroup, please feel free to absorb it into your own packages in any way you want under any GPL *,  BSD, LGPL * licenses.

One thing the developers of R might want to consider is to add the very basic optimization that Rigroup uses to base R so that it can better handle very very large datasets (ala BigData) more effieciently.  That is assuming there is not some other package that does this that I am unaware of or that a similar capability hasn't already been added to R-3.X since 2006.

The primary idea is simple and time worn ...  

Given a large unsorted vector of data whose elements have been assigned to n groups and whose group membership is represented by a second vector of equal length whoses values are members of {1,...,n}, you can easily calculate multiple group statistics for all n groups in just one pass through the unsorted data vector by using the group membership as an index into  one or multiple vectors of statistics such as count, sum, max, min, all, any, average, second moment, variance standard deviation, higher moments, etc ....  Since this is meant for very large vectors, it was implemented in C for speed.

For very very large unsorted data vectors, this one pass approach of using the group membership as an offset into potentially multiple "statistics vectors" is much much faster than trying to sort, or index, or subset the large unsorted data vector using the typical approaches of R and then calculating the stats and will work even if the group membership indicators do not span the set.  This is all that Rigroup did.  Pretty much just common sense to any old programmer who wanted to build portfolios and calculate basic portfolio stats but who has to deal with large amounts of data.    Perhaps sometime over the last 7 years, you have already added something similar, if so, please ignore this.

Given my withdrawl of this package I will also be removing myself from the R-devel mailing list so if you want to contact me for any reason please CC me directly (hopefully the sympatico.ca spam filter will not lose too many of you)!  It was an interesting 7 years.   Good luck to you all.

Take care,

Kevin