7 Mar
2011
7 Mar
'11
10:12 a.m.
There is a problem with the hash function currently used for floats. It multiplies by some number and casts the result to an UINT32. This will lead to many floats having the same hash value 0. Especially bad case:
hash_values(random(Float.MAX))
Also from those results I would say that random is broken for floats. It will always produce values lying within a few orders of magnitues in the above case. There is many options for good hash functions for floats/ints. Any preferences?
arne