介绍

数据归一化和标准化都是scaling，常用Normalization或Standardization表示。记录下R实现不同scaling方法。更多知识分享请到 https://zouhua.top/。

标准化R实现

  • Median scale normalization
  • Robust scale normalization
  • Unit scale normalization
  • z-scale normalization
  • Min-Max normalization 
  1. # method1: Median scale normalization 
  2. MDA_fun <- function(features){
  3.   # x for features X = (x1, x2, ..., xn)
  4.   value <- as.numeric(features)
  5.   d_mad <- mad(value)
  6.   x_scale <- (value - median(value))/d_mad
  8.   return(x_scale)  
  9. }
  10. dat_s1_MDA <- apply(dat, 1, MDA_fun)
  11. rownames(dat_s1_MDA) <- colnames(dat)
  13. # method2: Robust scale normalization 
  14. Robust_fun <- function(features){
  15.   # x for features X = (x1, x2, ..., xn)
  16.   value <- as.numeric(features)
  17.   q_value <- as.numeric(quantile(value))
  18.   remain_value <- value[value > q_value[2] & value < q_value[4]]
  20.   mean_value <- mean(remain_value)
  21.   sd_value <- sd(remain_value)
  23.   x_scale <- (value - mean_value)/sd_value
  25.   return(x_scale)  
  26. }
  28. # method3: Unit scale normalization 
  29. Unit_fun <- function(samples){
  30.   # v for samples v = (v1, v2, ..., vn)
  31.   value <- as.numeric(samples)
  32.   x_scale <- value / sqrt(sum(value^2))
  34.   return(x_scale)  
  35. }
  37. # method4: z-scale normalization  
  38. Zscore_fun <- function(features){
  39.   # x for features X = (x1, x2, ..., xn)
  40.   value <- as.numeric(features)
  41.   mean_value <- mean(value)
  42.   sd_value <- sd(value)
  44.   x_scale <- (value - mean_value)/sd_value
  46.   return(x_scale) 
  47. }
  49. # method5: Min-Max  normalization  
  50. Min_Max_fun <- function(features){
  51.   # x for features X = (x1, x2, ..., xn)
  52.   value <- as.numeric(features)
  53.   min_value <- min(value)
  54.   max_value <- max(value)
  56.   x_scale <- (value - min_value)/(max_value - min_value)
  58.   return(x_scale) 
  59. }

method1 2 4 5 的scaling的计算方式为减一个统计量再除以一个统计量，method3除以向量自身的长度，前者适合行向量，后者适合列向量，当然也不一定。

参考

  1. Data Normalization With R
  2. Median Absolute Deviation

参考文章如引起任何侵权问题，可以与我联系，谢谢。