analyse_xml.py 文件源码

def elbow_test(X, max_cluster):
    """
    This function performs the elbow test to determine the number of clusters for k-means clustering.

    Parameters
    ----------           
    X : numpy array
        2d list of floats.  

    max_cluster : int
        The maximum number of clusters to desirable.

    Returns
    -------
    number of clusters : int
        The number of clusters for kmeans clustering
    """
    from sklearn.cluster import KMeans
    from sklearn import metrics
    inertia_list = []
    s_list = []
    for cluster_cnt in range(max_cluster-1):
        k_means = KMeans(n_clusters=cluster_cnt+2)
        k_means.fit(X)
        k_means_labels = k_means.labels_
        s_factor = metrics.silhouette_score(X, k_means_labels, metric='euclidean')
        s_list.append(s_factor)
        kmeans_inertia = k_means.inertia_
        inertia_list.append(kmeans_inertia)

    inertia_cnt = 0
    i_diff_list = []
    for inertia in inertia_list:
        #look for the difference between each difference in cluster number
        if inertia_cnt != len(inertia_list) - 1:
            i_diff = inertia - inertia_list[inertia_cnt + 1]
            i_diff_list.append(i_diff)
        inertia_cnt = inertia_cnt + 1

    #find the biggest difference and use that number for the best number of cluster
    max_diff = max(i_diff_list)
    max_diff_index = i_diff_list.index(max_diff)
    #+3 because of the counting 
    best_no_cluster = max_diff_index + 3
    return best_no_cluster