Interesting problem. Sparse amend sorts the indices; they are mostly in
order but with some small values at the end. I previously implemented a
quicksort for integer lists, usually fast but this hit a worst-case.
Since I think the case is not unusual I have changed the choice of
pivot. Fixed for next beta.
Henry Rich
On 4/13/2021 7:52 PM, Ben Gorte wrote:
Perhaps in a bit of a clumsy way I tried to report last Thursday that
inserting a value in a large sparse array (changing an element from "zero"
into something else) had become quite slow.
SIZE=:1000000000000000
A =: 1 $. SIZE
i =: ?1000000 $ SIZE
v =: 1+?1000000 $ 1000
6!:2 'A =: v i } A' NB. insert 1 million values
0.179232
6!:2 ' A =: (999) 12345678901234 } A' NB. one more value
99.8308
The time depends strongly (quadratically?) on the number of elements
already in the array.
This is J-903. In J-902 the same amend takes 220s. Apparently it is a J9
thing, because I had not noticed it earlier. I checked with j64-807 and it
is only 0.24s.
Usually I'm on Linux, but meanwhile I tried the same in Windows 10 with
similar results.
Perhaps I'm the only one with this issue, but I wonder whether this
is this a glitch in the current J system that may be repaired -- or am I
somehow doing something inappropriate and is there a much better solution?
----------------------------------------------------------------------
For information about J forums see http://www.jsoftware.com/forums.htm
--
This email has been checked for viruses by AVG.
https://www.avg.com
----------------------------------------------------------------------
For information about J forums see http://www.jsoftware.com/forums.htm
