使用tslearn中的TimeSeriesKMeans时，将metric设置为softdtw后，聚类中心有误

Question

我在使用tslearn中的TimeSeriesKMeans时，将metric设置为softdtw时出现了问题，即聚类中心在可视化时非常离谱，不符实际。

但是将metric设置为欧式距离就没有问题，请教这是为什么？

def softdtw_KMeans(df_array, number_of_components):
    Kmeans = TimeSeriesKMeans(n_clusters=number_of_components,
                              metric='softdtw', init='random').fit(df_array)
    cluster_center = Kmeans.cluster_centers_
    return Kmeans, cluster_center

def cluster_center_plotter(df_array, number_of_components, time_series_class, cluster_center):
    global time
    cluster_10th_percentile = []
    cluster_90th_percentile = []

    for yi in range(number_of_components):
        for time in range(df_array.shape[1]):
            class_temporary = df_array[time_series_class == yi]
            number_1 = np.percentile(class_temporary[:, time], 10)
            number_2 = np.percentile(class_temporary[:, time], 90)
            cluster_10th_percentile.append(number_1)
            cluster_90th_percentile.append(number_2)
    cluster_10th_percentile = np.reshape(cluster_10th_percentile, (number_of_components, df_array.shape[1]))
    cluster_90th_percentile = np.reshape(cluster_90th_percentile, (number_of_components, df_array.shape[1]))

    plt.figure()
    for yi in range(number_of_components):
        plt.subplot(3, 3, yi + 1)
        for xx in range(3):
            plt.plot(cluster_10th_percentile[yi, :], "k-", alpha=.2)
            plt.plot(cluster_90th_percentile[yi, :], "k-", alpha=.2)
            plt.plot(cluster_center[yi].ravel(), "r-")
    return plt.show()

# softdtw_KMeans
km, cluster_center = KMeans_clustering.softdtw_KMeans(dataset, number_of_components)

# Plotting the clusters centre
# taking the 10th and 90th percentile to display along with the clusters centre
KMeans_clustering.cluster_center_plotter(dataset, number_of_components, time_series_class, cluster_center)

上图为使用欧式距离度量，下图为softdtw距离

Answer 1

你把报错信息发给我