Thanks for the advice. -----Original Message----- From: [email protected] [mailto:[email protected]] On Behalf Of Dan Bron Sent: December 3, 2009 11:41 AM To: Programming forum Subject: Re: [Jprogramming] averages of non-zero elements of matrices
Robert O'Boyle wrote: > The sparse array approach offers considerable potential That was a lark -- I do not suggest you pursue it. For one thing, even if you use sparse arrays, you won't reduce the number of "rectangular guards" very much. The short story is that you don't have to worry about the value of missing items, but you still need to worry about their indices. That is, the location of missing values is still there and will respond to inqueries; it is merely the value that is unspecified, and cannot play a role. The fact that in the two definitions of average, +/ remained the same but # turned into +/@:~:&0 exemplifies this difference. For another thing, sparse arrays don't get much use in J. Therefore they still have sharp corners. You will surely encounter phrases that should work, but don't. Also maybe the odd crash. So, if you do pursue this path, please report these problems as you encounter them (that'll help the next guy who uses sparse arrays). -Dan PS: No matter what approach you take (sparse or otherwise), to try to calculate with "missing values" is working against the grain of J. You're trying to force J to perform unnatural acts; expect it to balk. In the end, you'll have a much easier time if you use boxes, rank, and maybe cut, along the lines I sketched in my first message. ---------------------------------------------------------------------- For information about J forums see http://www.jsoftware.com/forums.htm ---------------------------------------------------------------------- For information about J forums see http://www.jsoftware.com/forums.htm
