IF I don't know the number of cluster

[shichenhu]

whether it's possible to use parament like SSE to determine the number of cluster,cause my data has no labels

[gsel9]

Hi!

You can try different number of clusters, measure the SSE for each run and select the number of clusters giving the smallest error. Maybe it could look something like this:

import numpy as np
from sklearn.datasets import load_iris

iris = load_iris(return_X_y=False)
X, y = iris.data, iris.target

clusters = [3, 6, 9]
results = {}

for num_clusters in clusters:

    dgufs = DGUFS(num_clusters=num_clusters)
    dgufs.fit(X)

    cluster_centers = [X[dgufs.memberships == i].mean(axis=0) 
                    for i in range(num_clusters)]

    clusterwise_sse = np.ones(num_clusters) * np.nan 
    for datapoint, label in zip(X, dgufs.memberships):
        clusterwise_sse[label] += np.square(datapoint - cluster_centers[label]).sum()

    results[f"num_clusters_{num_clusters}"] = clusterwise_sse

[shichenhu]

thank you so much, it's very useful

[gsel9]

Case closed