Question

我是R的新手，正在按照书中列出的算法编写我自己的k表示函数＆＃34; David Mackay的信息理论，推理和学习算法＆＃34; 。我的尝试如下。该函数采用1维数据。如何针对多变量数据（具有多个特征向量的数据）修改此问题？任何暗示都会有所帮助。提前致谢。

######generate data#####
x<-c(rnorm(500, 4,2), rnorm(500,5,1.5),rnorm(300,1,0.5))
data<-as.data.frame(x)
d_scaled<-as.data.frame(scale(data))

#### Kmeans algorithm ########

##### calculates the euclidean distance between two points p and q

    getEuclid_dist <- function(p,q) {
  dist <- 0.5*sum((p-q)^2) 
  return(dist)
}

##### finds the nearest mean to datapoints out of the set of means

get_nearest_mean <- function(data, means) {
  nmeans <- length(means$x)
  dist <- rep(NA,nmeans)
  for (k in 1:nmeans) {
    dist[k] <- getEuclid_dist(data,means$x[k])
  }
  minval <- which(dist %in% sort(dist)[1])
 return(minval)
}

mykmeans <- function( data, k ) {
  nd <- length(data$x)   # gets  total number of datapoints

  # 1. set each mean randomly to one of the datapoints

  m <- sample(1:nd,k)
  means <- list(x=data$x[m])
  mprev <- means

  # set initial responsibilities to 1
  # col = datapoints nd
  # row =  cluster
  # r(k,nd)=1 if it's in that cluster

  rold <- matrix(0,k,nd)
  rcurr <- matrix(1,k,nd)

  #2.  loop until the responsibilities have not changed

  while (!all(rcurr==rold)) {
    rold <- rcurr
    rcurr <- matrix(0,k,nd)

    #3. assignment step

    for( n in 1:nd) {
      kn <- get_nearest_mean(data$x[n],means)
      rcurr[kn,n] <- 1
    }

    # 4. recalculate the means

    for( nk in 1:k) {
      currx <- data$x[rcurr[nk,]==1]
      if (length(currx)!=0) {
        means$x[nk] <- mean(currx)
      }
     return(list(means$x, rcurr))

    }


  }
}
clusobj<-mykmeans(data=d_scaled,k=3)

如何修改我的kmeans函数来获取多元数据？

0 个答案: