helper.py 文件源码

def mean_shift(df, l1, l2, c1name, qt, cluster_all, bin_seeding):
    df1 = df.loc[df[c1name].isin([l1,l2])]
    pccols = [ i for i in range(0,50) ]
    xp = df1[pccols].as_matrix()
    bandwidth = 0
    if l1==l2:
        bandwidth = estimate_bandwidth(xp, quantile=qt)
    else:
        xp1 = df1.loc[df1[c1name]==l1, pccols].as_matrix()
        xp2 = df1.loc[df1[c1name]==l2, pccols].as_matrix()
        bandwidth1 = estimate_bandwidth(xp1, quantile=qt)
        bandwidth2 = estimate_bandwidth(xp2, quantile=qt)
        bandwidth = max(bandwidth1, bandwidth2)
    logging.info("compare (%d, %d) with width=%f", l1, l2, bandwidth)
    ms = MeanShift(bandwidth=bandwidth, cluster_all=cluster_all,
                   bin_seeding=bin_seeding)
    ms.fit(xp)        
    mslabels_unique = np.unique(ms.labels_)
    nc = len(mslabels_unique)
    nl = ms.labels_
    df.loc[df[c1name].isin([l1,l2]), c1name] = df.loc[df[c1name].isin([l1,l2]), c1name]*1000 +nl
    return nc, nl, bandwidth