数据分析：数据预处理--标准化初解释（一）

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

# method1: Median scale normalization 
MDA_fun <- function(features){
  # x for features X = (x1, x2, ..., xn)
  value <- as.numeric(features)
  d_mad <- mad(value)
  x_scale <- (value - median(value))/d_mad
  
  return(x_scale)  
}
dat_s1_MDA <- apply(dat, 1, MDA_fun)
rownames(dat_s1_MDA) <- colnames(dat)

# method2: Robust scale normalization 
Robust_fun <- function(features){
  # x for features X = (x1, x2, ..., xn)
  value <- as.numeric(features)
  q_value <- as.numeric(quantile(value))
  remain_value <- value[value > q_value[2] & value < q_value[4]]
  
  mean_value <- mean(remain_value)
  sd_value <- sd(remain_value)
  
  x_scale <- (value - mean_value)/sd_value
  
  return(x_scale)  
}

# method3: Unit scale normalization 
Unit_fun <- function(samples){
  # v for samples v = (v1, v2, ..., vn)
  value <- as.numeric(samples)
  x_scale <- value / sqrt(sum(value^2))
  
  return(x_scale)  
}

# method4: z-scale normalization  
Zscore_fun <- function(features){
  # x for features X = (x1, x2, ..., xn)
  value <- as.numeric(features)
  mean_value <- mean(value)
  sd_value <- sd(value)
  
  x_scale <- (value - mean_value)/sd_value
  
  return(x_scale) 
}

# method5: Min-Max  normalization  
Min_Max_fun <- function(features){
  # x for features X = (x1, x2, ..., xn)
  value <- as.numeric(features)
  min_value <- min(value)
  max_value <- max(value)
  
  x_scale <- (value - min_value)/(max_value - min_value)
  
  return(x_scale) 
}

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。