Welcome, guest | Sign In | My Account | Store | Cart

Space efficient, probabilistic set membership tester. Has no False Negatives but allows a rare False Positive.

Python, 111 lines
  1
  2
  3
  4
  5
  6
  7
  8
  9
 10
 11
 12
 13
 14
 15
 16
 17
 18
 19
 20
 21
 22
 23
 24
 25
 26
 27
 28
 29
 30
 31
 32
 33
 34
 35
 36
 37
 38
 39
 40
 41
 42
 43
 44
 45
 46
 47
 48
 49
 50
 51
 52
 53
 54
 55
 56
 57
 58
 59
 60
 61
 62
 63
 64
 65
 66
 67
 68
 69
 70
 71
 72
 73
 74
 75
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
from random import Random

class BloomFilter:
    # http://en.wikipedia.org/wiki/Bloom_filter

    def __init__(self, num_bytes, num_probes, iterable=()):
        self.array = bytearray(num_bytes)
        self.num_probes = num_probes
        self.num_bins = num_bytes * 8
        self.update(iterable)

    def get_probes(self, key):
        random = Random(key).random
        return (int(random() * self.num_bins) for _ in range(self.num_probes))

    def update(self, keys):
        for key in keys:
            for i in self.get_probes(key):
                self.array[i//8] |= 2 ** (i%8)

    def __contains__(self, key):
        return all(self.array[i//8] & (2 ** (i%8)) for i in self.get_probes(key))


##  Sample application  ##############################################

class SpellChecker(BloomFilter):

    def __init__(self, wordlistfiles, estimated_word_count=125000):
        num_probes = 14           # set higher for fewer false positives
        num_bytes = estimated_word_count * num_probes * 3 // 2 // 8
        wordlist = (w.strip() for f in wordlistfiles for w in open(f))
        BloomFilter.__init__(self, num_bytes, num_probes, wordlist)

    def find_misspellings(self, text):
        return [word for word in text.lower().split() if word not in self]


## Example of subclassing with faster probe functions ################

from hashlib import sha224, sha256

class BloomFilter_4k(BloomFilter):
    # 4Kb (2**15 bins) 13 probes. Holds 1,700 entries with 1 error per 10,000.

    def __init__(self, iterable=()):
        BloomFilter.__init__(self, 4 * 1024, 13, iterable)

    def get_probes(self, key):
        h = int(sha224(key.encode()).hexdigest(), 16)
        for _ in range(13):
            yield h & 32767     # 2 ** 15 - 1
            h >>= 15

class BloomFilter_32k(BloomFilter):
    # 32kb (2**18 bins), 13 probes. Holds 13,600 entries with 1 error per 10,000.

    def __init__(self, iterable=()):
        BloomFilter.__init__(self, 32 * 1024, 13, iterable)

    def get_probes(self, key):
        h = int(sha256(key.encode()).hexdigest(), 16)
        for _ in range(13):
            yield h & 262143    # 2 ** 18 - 1
            h >>= 18


if __name__ == '__main__':

    ## Compute effectiveness statistics for a 125 byte filter with 50 entries

    from random import sample
    from string import ascii_letters

    states = '''Alabama Alaska Arizona Arkansas California Colorado Connecticut
        Delaware Florida Georgia Hawaii Idaho Illinois Indiana Iowa Kansas
        Kentucky Louisiana Maine Maryland Massachusetts Michigan Minnesota
        Mississippi Missouri Montana Nebraska Nevada NewHampshire NewJersey
        NewMexico NewYork NorthCarolina NorthDakota Ohio Oklahoma Oregon
        Pennsylvania RhodeIsland SouthCarolina SouthDakota Tennessee Texas Utah
        Vermont Virginia Washington WestVirginia Wisconsin Wyoming'''.split()

    bf = BloomFilter(num_bytes=125, num_probes=14, iterable=states)

    m = sum(state in bf for state in states)
    print('%d true positives and %d false negatives out of %d positive trials'
          % (m, len(states)-m, len(states)))

    trials = 100000
    m = sum(''.join(sample(ascii_letters, 8)) in bf for i in range(trials))
    print('%d true negatives and %d false positives out of %d negative trials'
          % (trials-m, m, trials))

    c = ''.join(format(x, '08b') for x in bf.array)
    print('Bit density:', c.count('1') / float(len(c)))


    ## Demonstrate a simple spell checker using a 125,000 word English wordlist

    from glob import glob
    from pprint import pprint

    # Use the GNU ispell wordlist found at http://bit.ly/english_dictionary
    checker = SpellChecker(glob('/Users/raymondhettinger/dictionary/english.?'))
    pprint(checker.find_misspellings('''
        All the werldz a stage
        And all the mehn and wwomen merrely players
        They have their exits and their entrances
        And one man in his thaim pllays many parts
        His actts being sevven ages
    '''))

The get_probes() method generates a series of hashed lookups into the bitarray. There are many ways to do this and it is possible to get a much faster hashing function. The seeded random number generator approach was chosen for this recipe because 1) it produces as many uncorrelated bits as needed, 2) it is easy to understand, and 3) it works in both Python2 and Python3.

The example code shows the names of fifty US states stored in a 125-byte bloom filter using 14 probes. We verify that all 50 are found in the filter (guaranteed True Positives). We then try 100,000 random strings to see if they are in the filter. Typically, only a handful return a false positive.

Bloom filters are much more impressive with larger datasets. The above example also shows a list of all correctly spelled words in the English language compressed into a 330Kb bloom filter so that the spell checker can be kept entirely in memory.

For those interested in boosting performance, the above code includes example of BloomFilter subclasses with fast customized get_probes() methods. The first hardwires a 4Kb array, using 13 probes generated by the fast sha224 function. It performs well for up to 1,700 keys while keeping the false positive rate under 1 in 10,000. The second is a larger 32Kb array with 13 probes generated by sha256. It can hold up to 13,600 keys while keeping the false positive rate under 1 in 10,000. These arrays are compact enough to fit entirely in the L1 or L2 caches on modern processors (for example the Intel Sandy Bridge processors have a 32Kb L1 data cache and a 256Kb L2 cache).

The customized get_probes() methods can be further sped-up with traditional Python optimization techniques -- localize the global lookup of sha256() function, localize the builtin lookup for int(), and replace the range(13) call with a precomputed constant (or unroll the loop entirely):

def get_probes(self, key, sha256=sha256, int=int, range13=tuple(range(13))):
    h = int(sha256(key.encode()).hexdigest(), 16)
    for _ in range13:
        yield h & 262143    # 2 ** 18 - 1
        h >>= 18

1 comment

Manuel Garcia 12 years, 12 months ago  # | flag

Very minor point for the paranoid, if the keys are mostly sequential:

cryptohash_digest(key) :: from a python standard library cryptographic hash

random = Random(cryptohash_digest(key)).random