from tools import *

# https://en.wikipedia.org/wiki/Fuzzy_clustering

class FuzzyCMeans:

def __init__(self, n_clusters, initial_centers, data, max_iter=250, m=2, error=1e-5):

assert m > 1

#assert initial_centers.shape[0] == n_clusters

self.U = None

self.centers = initial_centers

self.max_iter = max_iter

self.m = m

self.error = error

self.data = data

def membership(self, data, centers):

U_temp = cdist(data, centers, 'euclidean')

U_temp = numpy.power(U_temp, 2/(self.m - 1))

denominator_ = U_temp.reshape(

(data.shape[0], 1, -1)).repeat(U_temp.shape[-1], axis=1)

denominator_ = U_temp[:, :, numpy.newaxis] / denominator_

return 1 / denominator_.sum(2)

def Centers(self, data, U):

um = U ** self.m

return (data.T @ um / numpy.sum(um, axis=0)).T

def newImage(self, U, centers, im):

best = numpy.argmax(self.U, axis=-1)

# print(best)

# numpy.round()

image = im.astype(int)

for i in range(256):

image = numpy.where(image == float(i), centers[best[i]][0], image)

return image

def compute(self):

self.U = self.membership(self.data, self.centers)

past_U = numpy.copy(self.U)

begin_time = datetime.datetime.now()

for i in range(self.max_iter):

self.centers = self.Centers(self.data, self.U)

self.U = self.membership(self.data, self.centers)

if norm(self.U - past_U) < self.error:

break

past_U = numpy.copy(self.U)

x = datetime.datetime.now() - begin_time

return self.centers, self.U, x

# that's how you run it, data being your data, and the other parameters being the basic FCM parameters such as numbe rof cluseters, degree of fuzziness and so on

# f = FuzzyCMeans(n_clusters=C, initial_centers=Initial_centers,

# data=data m=2, max_iter=1000, error=1e-5)

# centers, U, time = f.compute()