Today's fun and educational Python recipe

Terry Reedy tjreedy at udel.edu
Wed May 4 20:26:51 EDT 2011


On 5/4/2011 5:39 PM, Raymond Hettinger wrote:

> The 512 bits in h are progressively eaten-up between iterations.  So
> each pass yields a different (array index, bit_mask) pair.

Yeh, obvious now that I see it.

> It's easy to use the interactive prompt to show that different probes
> are produced on each pass:
>
>>>> bf = BloomFilter(num_bits=1000, num_probes=8)
>>>> pprint(list(bf.get_probes('Alabama')))
> [(19, 1073741824),
>   (11, 64),
>   (9, 134217728),
>   (25, 1024),
>   (24, 33554432),
>   (6, 16),
>   (7, 16777216),
>   (22, 1048576)]

Should have tried that.

> The 512 bits are uncorrelated -- otherwise sha512 wouldn't be much of
> a cryptographic hash ;)

> The fifty state example in the recipe is a reasonable demonstration
> that the recipe works as advertised.  It successfully finds all fifty
> states (the true positives) and it tries 100,000 negatives resulting
> in only a handful of false negatives.

I presume you mean 'false positives', as in the program comment and 
Wikipedia.

The test would be more convincing to many with 100000 other geographic 
names (hard to come by, I know), or other english names or words or even 
with longer random strings that matched the lengths of the state names. 
But an average of 5/100000 false positives in 5 runs is good.

-- 
Terry Jan Reedy




More information about the Python-list mailing list