analyze_clustering.py 文件源码

def get_k(clustering, depth = 10):
    """
    (ndarray, int) -> int
    clustering: ndarray -- linkage matrix representing hierarchical clustering
    depth: int -- the maximum depth to traverse clustering

    Returns the number of clusters to extract from the hierarchical clustering
    using the elbow method.
    """
    last = clustering[-depth: , 2]
    acceleration = np.diff(last, 2)
    acceleration_rev = acceleration[::-1]
    k = acceleration_rev.argmax() + 2
    return k