第3小問 Elbow method

k = 5

plt.figure(figsize=(10,5))

plt.plot()

plt.scatter(product_df[:, 0], product_df[:, 1], c='k')

plt.show()

# Fit clusters for various numbers of clusters

from sklearn.cluster import KMeans

K = range(1,10)

KM = [KMeans(n_clusters=k, random_state=0).fit(product_df) for k in K]

centroids = [km.cluster_centers_ for km in KM]? # cluster centroids

# Compute average euclidean distance between each point and its cluster centroid

from scipy.spatial.distance import cdist

D_k = [cdist(product_df, cent, 'euclidean') for cent in centroids]

cIdx = [np.argmin(D,axis=1) for D in D_k]

dist = [np.min(D,axis=1) for D in D_k]

avgWithinSS = [sum(d)/product_df.shape[0] for d in dist]

#Plot it

plt.plot()

plt.plot(K, avgWithinSS, 'b.-', ms=10)

plt.xlabel('Number of clusters')

plt.ylabel('Average within-cluster squared error')

plt.show()

SpectralClustering也是這樣做

把參數(shù)改改就好