dbscan.py 文件源码

def dbscan_partition(iterable, params, sample_weight=None):
    """
    :type iterable: iter
    :param iterable: iterator yielding ((key, partition), vector)
    :type params: dict
    :param params: dictionary containing sklearn DBSCAN parameters
    :rtype: iter
    :return: ((key, cluster_id), v)
    Performs a DBSCAN on a given partition of the data
    """
    # read iterable into local memory
    data = list(iterable)
    (key, part), vector = data[0]
    x = np.array([v for (_, __), v in data])
    y = np.array([k for (k, _), __ in data])
    # perform DBSCAN
    model = skc.DBSCAN(**params)
    # import sys
    # print(model, file=sys.stderr)
    weights = [sample_weight[k[0]] for k in x]
    c = model.fit_predict(x, sample_weight=weights)
    cores = set(model.core_sample_indices_)
    # yield (key, cluster_id), non-core samples labeled with *
    for i in xrange(len(c)):
        flag = '' if i in cores else '*'
        yield (y[i], '%i:%i%s' % (part, c[i], flag))