On Thu, Aug 06, 2020 at 08:50:27PM -0400, Christian Robert wrote:
∇z←(F fasteach) A;shp;vec;nub;res;idx
shp←⍴A ⍝ Shape and ...
vec←,A ⍝ ... ravel of arg.
nub←∪ vec ⍝ Vector of unique elements.
nub←∪ vec ⍝ Vector of unique elements.
res←F¨nub ⍝ Result for unique elts.
idx←nub ⍳ vec ⍝ Indices of arg in nub ...
z←shp ⍴ idx ⊃¨⊂ res ⍝ ... distribute result.
∇
The problem is indeed in iota and it reminds me a similar problem we had
with an unique before. For sufficiently big arrays, a different
algorithm is used where the left argument is sorted beforehand to speed
up searching; and something goes wrong there.
Smaller reproducer:
A←⌽⍳64
B←⍳6
B≡A[A⍳B]
0
Should be 1, of crouse.
-k
