sprand(m,n,density): for density > 0.5, the actual one is too big by 10% or more
Reported by Samuel GOUGEON (@sgougeon)
BUG DESCRIPTION:
----------------
When we use sprand(m,n,density) with a specified density, we expect that the output complies -- in average -- with the given density. But it's not the case. The density of the output clearly always overflows the given one. For instance, if we specify a density=0.9, the actual one will most often be 1.00 instead (no zeros!).
--> d0 = 0.9; clear d; for i=1:100, s = sprand(10,10, d0); d(i) = nnz(s)/100; end
--> mean(d)
ans =
0.9892
--> min(d)
ans =
0.96
--> stdev(d)
ans =
0.011693
ERROR LOG:
----------
None. The result does not comply with the specification.
HOW TO REPRODUCE THE BUG:
-------------------------
d0 = 0.9;
clear d
for i=1:100, s = sprand(10,10, d0); d(i) = nnz(s)/100; end
mean(d)
min(d)
stdev(d)
OTHER INFORMATION:
------------------
Same issue with Scilab .5.2 and with Scilab 6.0.1Activity
added 4.x and older Bugzilla Medium Sparse labels
By Samuel GOUGEON (@sgougeon)
Octave's output complies always with the given density (by 1 element):
d0 = 0.9; clear d for i=1:100, s = sprand(10,10, d0); d(i) = nnz(s)/100; end mean(d) ans = 0.90000 min(d) ans = 0.90000 std(d) ans = 7.8107e-016
-
By Samuel GOUGEON (@sgougeon)
Fix proposed for Scilab 6.0.2 in review at https://gitlab.com/scilab/legacy_codereview/-/wikis/changes/20612
Providing the exact density is possible (with grand(1,"prm",..) bu is 2-3 times slower.
With this fix, for instance:
Scilab 6.0.1: --> tic(); clear m, for i=1:100, s = sprand(1, 10000, 0.8); m(i)=nnz(s); end, [toc() mean(m) stdev(m)] ans = 0.487973 8787.01 29.708855
--> tic(); clear m, for i=1:100, s = sprand(1, 100000, 0.1); m(i)=nnz(s); end, [toc() mean(m) stdev(m)] ans = 1.1604901 10010.72 91.420229
--> tic(); clear m, for i=1:100, s = sprand(1, 1e6, 0.05); m(i)=nnz(s); end, [toc() mean(m) stdev(m)] ans = 4.4549477 50008.06 192.23254
With the patch:
--> tic(); clear m, for i=1:100, s = sprand(1, 10000, 0.8); m(i)=nnz(s); end, [toc() mean(m) stdev(m)] ans = 0.4361827 8000.8 13.070129
--> tic(); clear m, for i=1:100, s = sprand(1, 100000, 0.1); m(i)=nnz(s); end, [toc() mean(m) stdev(m)] ans = 0.9606055 10000.26 10.072084
--> tic(); clear m, for i=1:100, s = sprand(1, 1e6, 0.05); m(i)=nnz(s); end, [toc() mean(m) stdev(m)] ans = 3.1250801 50001.61 10.458089
assigned to @sgougeon