On Sat, Mar 19, 2011 at 8:10 AM, Nicolas M. Thiery <nicolas.thi...@u-psud.fr> wrote: > On Thu, Mar 17, 2011 at 03:29:08PM +0100, Jeroen Demeyer wrote: >> On 2011-03-17 15:04, Jason Grout wrote: >> > def hash(self): >> > h=0 >> > for i,j,entry in m.nonzero_entries(): # nonzero entries for sparse >> > matrices >> > h^=hash(entry)^i^j >> > return h >> >> Since you're only xorring, this will give a lot of collisions. I think >> something like >> >> def hash(self): >> n = m.ncols() >> h = (m.nrows() + m.ncols())**2 + n >> for i,j,entry in m.nonzero_entries(): >> h ^= hash(entry)*(i*n + j + 1) >> return h >> >> has less chance of collisions. Of course everything should be computed >> as a C long. > > Btw: any reason not to use Python's builtin hash function for tuples? > I assume it as been designed with precisely this kind of issues in > mind? Or do we want that a matrix gets the same hash in sparse and a > dense representation?
Equal sparse and dense matrices should have the same hash. Also, converting to a tuple and then hashing would be *way* slower for many matrices as one would have to construct a Python object for each entry. Not to say we can't do much better. - Robert -- To post to this group, send an email to sage-devel@googlegroups.com To unsubscribe from this group, send an email to sage-devel+unsubscr...@googlegroups.com For more options, visit this group at http://groups.google.com/group/sage-devel URL: http://www.sagemath.org
