예제 #1
0
파일: clustering.py 프로젝트: samyncn/bctpy
def agreement(ci, buffsz=None):
    '''
    Takes as input a set of vertex partitions CI of
    dimensions [vertex x partition]. Each column in CI contains the
    assignments of each vertex to a class/community/module. This function
    aggregates the partitions in CI into a square [vertex x vertex]
    agreement matrix D, whose elements indicate the number of times any two
    vertices were assigned to the same class.

    In the case that the number of nodes and partitions in CI is large
    (greater than ~1000 nodes or greater than ~1000 partitions), the script
    can be made faster by computing D in pieces. The optional input BUFFSZ
    determines the size of each piece. Trial and error has found that
    BUFFSZ ~ 150 works well.

    Parameters
    ----------
    ci : MxN np.ndarray
        set of M (possibly degenerate) partitions of N nodes
    buffsz : int | None
        sets buffer size. If not specified, defaults to 1000

    Returns
    -------
    D : NxN np.ndarray
        agreement matrix
    '''
    ci = np.array(ci)
    m, n = ci.shape

    if buffsz is None:
        buffsz = 1000

    if m <= buffsz:
        ind = dummyvar(ci)
        D = np.dot(ind, ind.T)
    else:
        a = np.arange(0, m, buffsz)
        b = np.arange(buffsz, m, buffsz)
        if len(a) != len(b):
            b = np.append(b, m)
        D = np.zeros((n,))
        for i, j in zip(a, b):
            y = ci[:, i:j + 1]
            ind = dummyvar(y)
            D += np.dot(ind, ind.T)

    np.fill_diagonal(D, 0)
    return D
예제 #2
0
def agreement(ci, buffsz=None):
    '''
    Takes as input a set of vertex partitions CI of
    dimensions [vertex x partition]. Each column in CI contains the
    assignments of each vertex to a class/community/module. This function
    aggregates the partitions in CI into a square [vertex x vertex]
    agreement matrix D, whose elements indicate the number of times any two
    vertices were assigned to the same class.

    In the case that the number of nodes and partitions in CI is large
    (greater than ~1000 nodes or greater than ~1000 partitions), the script
    can be made faster by computing D in pieces. The optional input BUFFSZ
    determines the size of each piece. Trial and error has found that
    BUFFSZ ~ 150 works well.

    Parameters
    ----------
    ci : MxN np.ndarray
        set of M (possibly degenerate) partitions of N nodes
    buffsz : int | None
        sets buffer size. If not specified, defaults to 1000

    Returns
    -------
    D : NxN np.ndarray
        agreement matrix
    '''
    ci = np.array(ci)
    m, n = ci.shape

    if buffsz is None:
        buffsz = 1000

    if m <= buffsz:
        ind = dummyvar(ci)
        D = np.dot(ind, ind.T)
    else:
        a = np.arange(0, m, buffsz)
        b = np.arange(buffsz, m, buffsz)
        if len(a) != len(b):
            b = np.append(b, m)
        D = np.zeros((n, ))
        for i, j in zip(a, b):
            y = ci[:, i:j + 1]
            ind = dummyvar(y)
            D += np.dot(ind, ind.T)

    np.fill_diagonal(D, 0)
    return D
예제 #3
0
def agreement_weighted(ci, wts):
    '''
    D = AGREEMENT_WEIGHTED(CI,WTS) is identical to AGREEMENT, with the
    exception that each partitions contribution is weighted according to
    the corresponding scalar value stored in the vector WTS. As an example,
    suppose CI contained partitions obtained using some heuristic for
    maximizing modularity. A possible choice for WTS might be the Q metric
    (Newman's modularity score). Such a choice would add more weight to
    higher modularity partitions.

    NOTE: Unlike AGREEMENT, this script does not have the input argument
    BUFFSZ.

    Parameters
    ----------
    ci : MxN np.ndarray
        set of M (possibly degenerate) partitions of N nodes
    wts : Mx1 np.ndarray
        relative weight of each partition

    Returns
    -------
    D : NxN np.ndarray
        weighted agreement matrix
    '''
    ci = np.array(ci)
    m, n = ci.shape
    wts = np.array(wts) / np.sum(wts)

    D = np.zeros((n, n))
    for i in range(m):
        d = dummyvar(ci[i, :].reshape(1, n))
        D += np.dot(d, d.T) * wts[i]
    return D
예제 #4
0
파일: clustering.py 프로젝트: samyncn/bctpy
def agreement_weighted(ci, wts):
    '''
    D = AGREEMENT_WEIGHTED(CI,WTS) is identical to AGREEMENT, with the
    exception that each partitions contribution is weighted according to
    the corresponding scalar value stored in the vector WTS. As an example,
    suppose CI contained partitions obtained using some heuristic for
    maximizing modularity. A possible choice for WTS might be the Q metric
    (Newman's modularity score). Such a choice would add more weight to
    higher modularity partitions.

    NOTE: Unlike AGREEMENT, this script does not have the input argument
    BUFFSZ.

    Parameters
    ----------
    ci : MxN np.ndarray
        set of M (possibly degenerate) partitions of N nodes
    wts : Mx1 np.ndarray
        relative weight of each partition

    Returns
    -------
    D : NxN np.ndarray
        weighted agreement matrix
    '''
    ci = np.array(ci)
    m, n = ci.shape
    wts = np.array(wts) / np.sum(wts)

    D = np.zeros((n, n))
    for i in range(m):
        d = dummyvar(ci[i, :].reshape(1, n))
        D += np.dot(d, d.T) * wts[i]
    return D