Details
-
Bug
-
Status: Closed
-
Major
-
Resolution: Fixed
-
None
-
None
-
None
Description
Relative to the blocksize, if generated matrices get very sparse, we see a systematic bias toward the upper end of the value range.
X =rand(rows=10000000,cols=100000,min =0,max =100,sparsity=0.00000001 ) write(X, "tmp/sparse.txt", format="text")
After closer investigation, this was due to correlated nnz and value random number generators (same seed). We now add a prime to the second random number generator, which fixed the problem