D-Net

def sorensendice(*args):

    """

    .. function:: sorensendice(jpack1,jpack2)

    Return jaccard similarity value of two jpacks.

    Example:

    >>> table1('''

    ... user1   movie1 20

    ... user1   movie2 30

    ... user2   movie1 40

    ... user2   movie3 90

    ... user2   movie4 90

    ... user3   movie1 40

    ... user3   movie3 80

    ... user4   movie1 70

    ... user4   movie2 10

    ... ''')

    NOTE that only column b is jgrouped because *jaccard* operates on packs as sets, not weighted values, So packing

    also column c would not make any difference.

    >>> sql(\"""select u1.userid,u2.userid, sorensendice(u1.pk, u2.pk) as similarity

    ...     from

    ...         (select a as userid, jgroup(b)  as pk from table1 group by a) as u1,

    ...         (select a as userid, jgroup(b) as pk from table1 group by a) as u2

    ...     where u1.userid<u2.userid\""")

    userid | userid | similarity

    ----------------------------

    user1  | user2  | 0.4

    user1  | user3  | 0.5

    user1  | user4  | 1.0

    user2  | user3  | 0.8

    user2  | user4  | 0.4

    user3  | user4  | 0.5

    """

    if len(args)!=2:

        raise functions.OperatorError("sorensendice","operator takes exactly two arguments")

    try:

        r=jopts.fromj(args[0])

        s=jopts.fromj(args[1])

    except Exception,e:

        raise functions.OperatorError("sorensendice"," Wrong format arguments: %s" %(e))

    rset=set([tuple(x) if type(x)==list else x for x in r])

    sset=set([tuple(x) if type(x)==list else x for x in s])

    return 2 * float(len( rset & sset ))/(len(rset) + len(sset) )

sorensendice.registered=True