def __init__(self, X, ct_indices=[1, 4, 6]):

# def __init__(self, X, ct_indices=[1, 4, 7]):

# def __init__(self, X, ct_indices=[2, 4, 7]):

self.X = X

self.cts = X[ct_indices]

self.radius = np.zeros(self.cts.shape[0])

self.ct_indices = np.zeros(self.X.shape[0])

self.clusters = []

self.fig, self.ax = plt.subplots()

self.colors = ['#9ACD32', '#6B8E23', '#EE82EE']

def fit(self, iter_times=4):

for i in range(iter_times):

self.ax.cla()

self.ax.xaxis.set_major_locator(MultipleLocator(2.0))

self.ax.yaxis.set_major_locator(MultipleLocator(2.0))

self.calculate_radius()

self.clusters.append(self.ct_indices)

pca_X = pca(self.X)

pca_cts = pca(self.cts)

self.ax.set_title(f"KMeans: Step {i + 1}")

self.ax.scatter(pca_X[:, 0], pca_X[:, 1], c=[self.colors[i] for i in self.ct_indices], s=80)

for l in range(self.X.shape[0]):

self.ax.text(pca_X[l, 0] + 0.08, pca_X[l, 1], l + 1, verticalalignment='center', horizontalalignment='left')

self.ax.scatter(pca_cts[:, 0], pca_cts[:, 1], c='r', s=100)

for l in range(self.cts.shape[0]):

cir = Circle(xy=(pca_cts[l, 0], pca_cts[l, 1]), radius=self.radius[l], alpha=0.5)

self.ax.add_patch(cir)

plt.savefig(f"kmeans_{i}.png")

plt.pause(3)

self.new_cts()

plt.ioff()

plt.show()

print(self.clusters)

def calculate_radius(self):

dist = np.sqrt(((self.X - self.cts[:, np.newaxis])**2).sum(axis=2))

self.ct_indices = np.argmin(dist, axis=0)

dist = dist.min(axis=0)

tmp = []

for i in range(self.cts.shape[0]):

if np.sum(self.ct_indices == i) == 0:

tmp.append(0)

else:

tmp.append(dist[self.ct_indices == i].max(axis=0))

self.radius = np.array(tmp)

def new_cts(self):

tmp = []

for i in range(self.cts.shape[0]):

if np.sum(self.ct_indices == i) == 0:

tmp.append(self.cts[i])

else:

tmp.append(self.X[self.ct_indices == i].mean(axis=0))