I just had to set up a Bloom filter. I found a seemingly cogent implementation by one Bruno Martins, but it has a design decision that I found difficult to swallow.
To get the multiple hashes, he: 1: used Rabin fingerprints to get an hash. 2: rehashed that hash with varying seeds to get the multiple hashes. In bear-of-little-brain mode, this struck me as approximately nuts. The initial hash already threw out a ton of information. So how can messing with it further give truly distinctive slices?
