Describe the workflow you want to enable

I would like a new class: sklearn.cluster.KMedians (or an option to sklearn.cluster.KMeans) that allows the methods to use medians instead of means.

K-n clustering can greatly improve some instances where there are outliers. Using the median minimizes outliers.

Describe your proposed solution

Create a new class sklearn.cluster.KMedians that works the same as sklearn.cluster.KMeans but instead uses median to compute the new centroids (instead of using mean)

Describe alternatives you've considered, if relevant

Using a version I wrote:

class mu_type(enum.IntEnum):
    mean = 1
    median = 2


#
# Euclidean distance between two vectors
#
def euclidean_distance(row1, row2):
    distance = 0.0
    for i in range(len(row1)):
        distance += (row1[i] - row2[i])**2
    return math.sqrt(distance)


#
# k-means and k-medians clustering
# mu defines if the algorithm runs as k-means or k-medians
#
def k_m_clustering_2(x, c, mu=mu_type.mean):
    
    d = [[0]*len(c) for i1 in range(len(x))]
    
    l = [0]*len(x)
    c_last = [0]*len(c)
    
    while c_last != c:
        
        # Save the last list of center points to compare for updates later
        c_last = c.copy()
        
        for i in range(len(x)):
            for j in range(len(c)):
                if DEBUG:
                    print('distance between: ', end='')
                    print(x[i], end='')
                    print(' and ', end='')
                    print(c[j], end='')
                    
                d[i][j] = euclidean_distance(x[i], c[j])
                
                if DEBUG:
                    print(' = ', end='')
                    print(d[i][j])
                    
            l[i] = d[i].index(min(d[i]))

        if DEBUG:
            print("L = ", end='')
            print(l)
            print()
            print()
        
        if DEBUG:
            print("Old center points: ", end='')
            print(c)
            print()

        #
        # Update center points
        #
        for i in range(len(c)):
    
            #
            # Compute mean or median for new center points
            #       
            if mu == mu_type.mean:
                count = 0
                _x = []
                _y = []
                for j in range(len(l)):
                    if l[j] == i:
                        count += 1
                        _x.append(x[j][0])
                        _y.append(x[j][1])
                c[i] = [(sum(_x)/count), (sum(_y)/count)]
                
            elif mu == mu_type.median:
                x_y = []
                for j in range(len(l)):
                    if l[j] == i:
                        x_y.append([x[j][0], x[j][1]])
                c[i] = median(x_y)
            
        if DEBUG:
            print("New center points: ", end='')
            print(c)
            print()
    
    if DEBUG:
        print("Center points have not changed\n")
        print("Final: ", end='')
        print(l)
        print()
    
    return c, l




# Call it:

DEBUG=True

c = [[2, 2], [3, 4], [6, 2]]
x = [[1, 2], [2, 1], [1, 3], [5, 4], [6, 3], [7, 2], [6, 1]]

c_arr, l_arr = k_m_clustering_2(x, c, mu_type.median)

for i in range(len(x)):
    print('x' + str(i+1) + ' = ' + str(l_arr[i]))
print('\n')

Output:

distance between: [1, 2] and [2, 2] = 1.0
distance between: [1, 2] and [3, 4] = 2.8284271247461903
distance between: [1, 2] and [6, 2] = 5.0
distance between: [2, 1] and [2, 2] = 1.0
distance between: [2, 1] and [3, 4] = 3.1622776601683795
distance between: [2, 1] and [6, 2] = 4.123105625617661
distance between: [1, 3] and [2, 2] = 1.4142135623730951
distance between: [1, 3] and [3, 4] = 2.23606797749979
distance between: [1, 3] and [6, 2] = 5.0990195135927845
distance between: [5, 4] and [2, 2] = 3.605551275463989
distance between: [5, 4] and [3, 4] = 2.0
distance between: [5, 4] and [6, 2] = 2.23606797749979
distance between: [6, 3] and [2, 2] = 4.123105625617661
distance between: [6, 3] and [3, 4] = 3.1622776601683795
distance between: [6, 3] and [6, 2] = 1.0
distance between: [7, 2] and [2, 2] = 5.0
distance between: [7, 2] and [3, 4] = 4.47213595499958
distance between: [7, 2] and [6, 2] = 1.0
distance between: [6, 1] and [2, 2] = 4.123105625617661
distance between: [6, 1] and [3, 4] = 4.242640687119285
distance between: [6, 1] and [6, 2] = 1.0
L = [0, 0, 0, 1, 2, 2, 2]

Old center points: [[2, 2], [3, 4], [6, 2]]

New center points: [[1, 3], [5, 4], [6, 3]]

distance between: [1, 2] and [1, 3] = 1.0
distance between: [1, 2] and [5, 4] = 4.47213595499958
distance between: [1, 2] and [6, 3] = 5.0990195135927845
distance between: [2, 1] and [1, 3] = 2.23606797749979
distance between: [2, 1] and [5, 4] = 4.242640687119285
distance between: [2, 1] and [6, 3] = 4.47213595499958
distance between: [1, 3] and [1, 3] = 0.0
distance between: [1, 3] and [5, 4] = 4.123105625617661
distance between: [1, 3] and [6, 3] = 5.0
distance between: [5, 4] and [1, 3] = 4.123105625617661
distance between: [5, 4] and [5, 4] = 0.0
distance between: [5, 4] and [6, 3] = 1.4142135623730951
distance between: [6, 3] and [1, 3] = 5.0
distance between: [6, 3] and [5, 4] = 1.4142135623730951
distance between: [6, 3] and [6, 3] = 0.0
distance between: [7, 2] and [1, 3] = 6.082762530298219
distance between: [7, 2] and [5, 4] = 2.8284271247461903
distance between: [7, 2] and [6, 3] = 1.4142135623730951
distance between: [6, 1] and [1, 3] = 5.385164807134504
distance between: [6, 1] and [5, 4] = 3.1622776601683795
distance between: [6, 1] and [6, 3] = 2.0
L = [0, 0, 0, 1, 2, 2, 2]

Old center points: [[1, 3], [5, 4], [6, 3]]

New center points: [[1, 3], [5, 4], [6, 3]]

Center points have not changed

Final: [0, 0, 0, 1, 2, 2, 2]

x1 = 0
x2 = 0
x3 = 0
x4 = 1
x5 = 2
x6 = 2
x7 = 2

Additional context

No response