EM Clustering Demo

EM Clustering Demo
by Alissa Liu
for Game Maker 8.0

Description:
Demonstrates data clustering/grouping using Expectation–maximization method. In this demonstration, n two-dimensional vector data (represented by points) are given in random (with specific range) and also given specific number of clusters (k, max=7). With given random parameters of mixture model of normal distributions (means of each component/cluster represented by X-mark points and std. devs represented by ellipses), the algorithm attempts to perform clustering/grouping data until convergence, i.e. log-likelihood of the mixture parameters reaches optimum value. Colors of each data indicate their cluster/group.

Link: [You must be registered and logged in to see this link.]

Screenshots:

Spoiler:

Spoiler:

Spoiler:

Implementation usage:
The function is EM_cluster(). Requires the following input arguments:
- n: number of data
- k: number of clusters
- cx[0..k-1], cy[0..k-1]: mean value of each mixture component (as cluster)
- sx[0..k-1], sy[0..k-1]: Std. deviation value of each mixture component
- xp[0..n-1], yp[0..n-1]: data vector
- cp[0..n-1]: cluster number assigned on data or latent variables of each data (-1 as uninitialized)

The resulting output is cp[0..n-1] value of each data

Note:
Basically, EM algorithm does similar process as [You must be registered and logged in to see this link.] and because k-means is a variant of generalized EM algorithm. However EM using mixture model of normal distributions can give better results.

Asuna kembali menyerang dengan algoritma2 keren
walau skrg msih gk tau mau di pake buat apaan, tpi tetep hrus di download dulu gmk ny

nice work

ini seperti di example yg sebelumnya: [You must be registered and logged in to see this link.] , tapi dgn algoritma yg berbeda

kalo belum tau, ini merupakan bagian dari AI (aspek Data mining)
pada penerapannya mungkin contohnya saat objek2nya ingin mencari & membentuk kelompok berdasarkan kedekatan posisinya

iy, pas di lihat hasil ny sama dgn yg sebelum ny
ya, I see

Asuna strikes again

(sementara gw kejebak internet lelet )


Ok asuna thanks for sharing, will DEFINITELY try this. Gw yakin ini top punya