You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
When experimenting with reenabling dune runtest under Windows (MinGW) in #292 I've noticed that they fail expect tests due to different stddev's being printed.
There's a reported difference with both OCaml 4.14, 5.0, and 5.2:
stats dist:
- num: 100000, avg: -7215552342607541.00, stddev: 2666234426234218496.00, median -16620417636667326, min -4611371852367564818, max 4611613630315464842+ num: 100000, avg: -7215552342607541.00, stddev: 2666234426234218500.00, median -16620417636667326, min -4611371852367564818, max 4611613630315464842
Stddev is computed as follows:
let stddev =Hashtbl.fold
(funiresm -> m +. (float_of_int i -.!avg) **2.*. float_of_int res)
tbl 0.|> (funs -> if!num>0then s /. float_of_int !num else s)
|> sqrt
and printed as:
Printf.bprintf out
" num: %d, avg: %.2f, stddev: %.2f, median %d, min %d, max %d\n"!num !avg stddev !median min_idx max_idx;
I've not dug further into the reason at this point, but suspect that one of the involved floating point operations is causing the difference.
When experimenting with reenabling
dune runtest
under Windows (MinGW) in #292 I've noticed that they fail expect tests due to different stddev's being printed.There's a reported difference with both OCaml 4.14, 5.0, and 5.2:
Stddev is computed as follows:
and printed as:
I've not dug further into the reason at this point, but suspect that one of the involved floating point operations is causing the difference.
Full details from the 5.2 run:
https://github.com/c-cube/qcheck/actions/runs/9796388729/job/27050729649?pr=292
The text was updated successfully, but these errors were encountered: