scale函数注意事项

Scale使用过程中需要注意以下几点：

1. scale后接向量时，输出结果为array，注意不是向量

   > test <- scale(c(1,2,3))
   > test
        [,1]
   [1,]   -1
   [2,]    0
   [3,]    1
   attr(,"scaled:center")
   [1] 2
   attr(,"scaled:scale")
   [1] 1
   > is.array(test)
   [1] TRUE
   

2. scale后接矩阵

   若输入为矩阵，则以列为基准去标准化，生成的结果是一个array，行列与原matrix一致

   > test
        a  b  c
   [1,] 1  1 12
   [2,] 3  6 11
   [3,] 9 10  9
   > test2 <- scale(test)
   > test2
                 a           b          c
   [1,] -0.8006408 -1.03490978  0.8728716
   [2,] -0.3202563  0.07392213  0.2182179
   [3,]  1.1208971  0.96098765 -1.0910895
   attr(,"scaled:center")
           a         b         c 
    4.333333  5.666667 10.666667 
   attr(,"scaled:scale")
          a        b        c 
   4.163332 4.509250 1.527525 
   > is.array(test2)
   [1] TRUE
   

3. scale后接dataframe

   > test <- data.frame(a=c(1,3,9),b=c(1,6,10),c=c(12,11,9),d=c(100,2,3))
   > test
     a  b  c   d
   1 1  1 12 100
   2 3  6 11   2
   3 9 10  9   3
   > scale(test) #默认以列为基准进行标准化
   #等同于 scale(as.matrix(test)) 以及 apply(test,2,scale)
                 a           b          c          d
   [1,] -0.8006408 -1.03490978  0.8728716  1.1546550
   [2,] -0.3202563  0.07392213  0.2182179 -0.5862095
   [3,]  1.1208971  0.96098765 -1.0910895 -0.5684455
   attr(,"scaled:center")
           a         b         c         d 
    4.333333  5.666667 10.666667 35.000000 
   attr(,"scaled:scale")
           a         b         c         d 
    4.163332  4.509250  1.527525 56.293872
   > scale(as.matrix(test))
                 a           b          c          d
   [1,] -0.8006408 -1.03490978  0.8728716  1.1546550
   [2,] -0.3202563  0.07392213  0.2182179 -0.5862095
   [3,]  1.1208971  0.96098765 -1.0910895 -0.5684455
   attr(,"scaled:center")
           a         b         c         d 
    4.333333  5.666667 10.666667 35.000000 
   attr(,"scaled:scale")
           a         b         c         d 
    4.163332  4.509250  1.527525 56.293872
   > apply(test,2,scale) # 2 代表依据列执行
                 a           b          c          d
   [1,] -0.8006408 -1.03490978  0.8728716  1.1546550
   [2,] -0.3202563  0.07392213  0.2182179 -0.5862095
   [3,]  1.1208971  0.96098765 -1.0910895 -0.5684455
   
   > apply(test,1,scale) # 1 代表依据行执行，最终结果为何是这样，因为第一行scale后得到的数据放到第一列了(类似scale后接向量的感觉)，第二行scale后放第二列了，等等以此类推
   # 等同于scale(t(as.matrix(test)))
              [,1]       [,2]       [,3]
   [1,] -0.5735393 -0.6185896  0.3904344
   [2,] -0.5735393  0.1237179  0.7027819
   [3,] -0.3441236  1.3608971  0.3904344
   [4,]  1.4912023 -0.8660254 -1.4836507
   > scale(t(as.matrix(test)))
           [,1]       [,2]       [,3]
   a -0.5735393 -0.6185896  0.3904344
   b -0.5735393  0.1237179  0.7027819
   c -0.3441236  1.3608971  0.3904344
   d  1.4912023 -0.8660254 -1.4836507
   attr(,"scaled:center")
   [1] 28.50  5.50  7.75
   attr(,"scaled:scale")
   [1] 47.947888  4.041452  3.201562
   
   #所以，如果想按行scale怎么处理
   > t(apply(test,1,scale))
              [,1]       [,2]       [,3]       [,4]
   [1,] -0.5735393 -0.5735393 -0.3441236  1.4912023
   [2,] -0.6185896  0.1237179  1.3608971 -0.8660254
   [3,]  0.3904344  0.7027819  0.3904344 -1.4836507