OGD {ORKM}

Online gradient descent algorithm for online single-view data clustering

Description

Online gradient descent is an optimisation algorithm in machine learning for when the amount of data is too large to process all the data at the same time. In this algorithm, the model parameters are updated based on a single training sample, rather than using the entire training set. The direction of each update is determined by the direction of the gradient of the current sample, and the local or global extremes of the gradient descent algorithm depend on the order of the sampled samples. Compared to Batch Gradient Descent (BGD) algorithm, online gradient descent algorithms can process data streams and update the model as they process the data, and are therefore more efficient for large-scale data. However, online gradient descent algorithm should only be used if the data stream is continuously present and updated.

Usage

OGD(X, K, gamma, max.m, chushi, yita, epsilon, truere, method = 0)

Arguments

X

data matrix

K

number of cluster

gamma

step size

yita

the regularized parameter

truere

true cluster result

max.m

max iter

epsilon

epsilon

chushi

the initial value

method

caculate the index of NMI

Value

result,NMI,M

Author(s)

Miao Yu

Examples

yita=0.5;V=2;K=3;chushi=100;epsilon=1;gamma=0.1;max.m=10;n1=n2=n3=70
 X1<-rnorm(n1,20,2);X2<-rnorm(n2,25,1.5);X3<-rnorm(n3,30,2) 
 Xv<-c(X1,X2,X3)
 data<-matrix(Xv,n1+n2+n3,2)
 data[1:70,2]<-1;data[71:140,2]<-2;data[141:210,2]<-3
 X<-matrix(data[,1],n1+n2+n3,1) 
 truere=data[,2]
 lamda1<-0.2;lamda2<-0.8
 lamda<-matrix(c(lamda1,lamda2),nrow=1,ncol=2)
 sol.svd <- svd(lamda)
  U1<-sol.svd$u
 D1<-sol.svd$d
 V1<-sol.svd$v
 C1<-t(U1)
 Y1<-C1/D1
 view<-V1
 view1<-matrix(view[1,])
 view2<-matrix(view[2,])
 X1<-matrix(view1,n1+n2+n3,1)
 X2<-matrix(view2,n1+n2+n3,1)
 OGD(X=X1,K=K,gamma=gamma,max.m=max.m,chushi=chushi,
yita=yita,epsilon=epsilon,truere=truere,method=0)