Make LAGraph_random configurable with ranges

[szarnyasg]

Followup of issue Graphegon/pygraphblas#8. LAGraph_random fills zero-values in nonzero positions as demonstrated by the following code:

int seed = 42;
GrB_Index nvals;
LAGraph_random(&A, GrB_BOOL, 10, 10, 4, false, true, false, false, true, &seed);
GrB_Matrix_nvals(A, &nvals);
GrB_wait();
GxB_print(A, GxB_SHORT);

  row: 0 : 1 entries [0:0]
    column 9:   1
  row: 1 : 1 entries [1:1]
    column 4:   1
  row: 2 : 1 entries [2:2]
    column 6:   1
  row: 3 : 1 entries [3:3]
    column 4:   0
  row: 4 : 2 entries [4:5]
    column 1:   1
    column 3:   0
  row: 6 : 1 entries [6:6]
    column 2:   1
  row: 9 : 1 entries [7:7]
    column 0:   1

I'm not sure whether this is the intended behaviour but based on the random generators for INT and FP types, putting "explicit zeros" is intentional. So maybe this should be kept as is but documented in the header comment of LAGraph_random. What do you think?

PS: For booleans, setting the make_pattern parameter to true works around the problem.

[DrTimothyAldenDavis]

Yes, it's intentional since explicit zeros are sometimes interesting. So the feature should be kept, but we could add functionality. For this case, and perhaps other uses (like int8 for example) it might be nice to be able to specify a range of values that the entries can be given. I do this in my GrB.random function in my MATLAB interface, where a range can be specified (lo and hi).

…

On Tue, Jan 21, 2020 at 1:32 PM Gabor Szarnyas ***@***.***> wrote: Followup of issue Graphegon/pygraphblas#8 <Graphegon/pygraphblas#8>. LAGraph_random <https://github.com/GraphBLAS/LAGraph/blob/master/Source/Utility/LAGraph_random.c> fills zero-values in nonzero positions as demonstrated by the following code: int seed = 42; GrB_Index nvals;LAGraph_random(&A, GrB_BOOL, 10, 10, 4, false, true, false, false, true, &seed);GrB_Matrix_nvals(A, &nvals);GrB_wait();GxB_print(A, GxB_SHORT); row: 0 : 1 entries [0:0] column 9: 1 row: 1 : 1 entries [1:1] column 4: 1 row: 2 : 1 entries [2:2] column 6: 1 row: 3 : 1 entries [3:3] column 4: 0 row: 4 : 2 entries [4:5] column 1: 1 column 3: 0 row: 6 : 1 entries [6:6] column 2: 1 row: 9 : 1 entries [7:7] column 0: 1 I'm not sure whether this is the intended behaviour but based on the random generators for INT and FP types, putting "explicit zeros" is intentional. So maybe this should be kept as is but documented in the header comment of LAGraph_random. What do you think? PS: For booleans, setting the make_pattern parameter to true works around the problem. — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#59?email_source=notifications&email_token=AEYIIONSNLCC23RMKGSUT3LQ65EVJA5CNFSM4KJY33HKYY3PNVWWK3TUL52HS4DFUVEXG43VMWVGG33NNVSW45C7NFSM4IHXL7DQ>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AEYIIOMA7GHKCXJ3SGVTVBTQ65EVJANCNFSM4KJY33HA> .