It's comparison tolerance that I think you would want to mess with, to minimize the occurrences of this issue.
https://www.jsoftware.com/help/dictionary/dx009.htm But you're not going to be able to set it higher than 9!:19]1e_12 If you are not concerned with floating point, though, you might want to try runiform=: {{ rnd=. ?@#&0 relist=. (-/ , {:)@|. x:> (rnd y) #.each <relist x }} Also, testing this out, I noticed that my rephrasing of leftR and rightR hit a length error. Oops... I should have noticed that earlier. There's a variety of ways of fixing that, including: runiform=: {{ rnd=. ($~ $-.1:) ?@#&0 relist=. (-/ , {:)@|. x:> (rnd y) #.each <relist x }} FYI, -- Raul On Mon, Jan 10, 2022 at 7:03 AM Pawel Jakubas <jakubas.pa...@gmail.com> wrote: > > Thanks Henry and Raul for your responses. > Indeed, I have encountered limitations of floating point arithmetic. > > showmismatch is great idea to quickly detect cases validating property. > > Usually I will want to have something like this though (100 runs or so): > run=: 3 : 0 > shape=.1+?2#100 > m=.genUniformMatrix shape > s=. _100 100 runiform 2 > s checkEqOfMatricesWithScalars m > ) > (+/)(run"0)100#0 > 100 > > I will try to experiment also with digit precision and rounding as a > measure to avoid those limitations. > > Cheers, > > Pawel Jakubas > ---------------------------------------------------------------------- > For information about J forums see http://www.jsoftware.com/forums.htm ---------------------------------------------------------------------- For information about J forums see http://www.jsoftware.com/forums.htm
