Question

我正在重写scale函数。我想获得数据集树的相同结果。但是当我运行代码时，我得到一个非常奇怪的结果。

z_function = function(x){
    (x - mean(x))/sd(x)}


scale_function = function(x){
    result = apply(x,2,z_function)
    att_mean = apply(x,2,mean)
    att_sd = apply(x,2,sd)
    attributes(result) = list("scaled:center" = att_mean,"scaled:scale"= att_sd)
    result
}

scale_function(trees)

预期结果：

        #             Girth     Height      Volume
        # [1,] -1.57685421 -0.9416472 -1.20885469
        # [2,] -1.48125614 -1.7263533 -1.20885469
        # [3,] -1.41752409 -2.0402357 -1.21493821
        # [4,] -0.87580169 -0.6277648 -0.83775985
        # [5,] -0.81206964  0.7847060 -0.69175532
        # [6,] -0.78020362  1.0985884 -0.63700362
        # [7,] -0.71647157 -1.5694121 -0.88642802
        # [8,] -0.71647157 -0.1569412 -0.72825645
        # [9,] -0.68460554  0.6277648 -0.46058149
        # [10,] -0.65273952 -0.1569412 -0.62483658
        # [11,] -0.62087350  0.4708236 -0.36324513
        # [12,] -0.58900747  0.0000000 -0.55791784
        # [13,] -0.58900747  0.0000000 -0.53358375
        # [14,] -0.49340940 -1.0985884 -0.53966727
        # [15,] -0.39781133 -0.1569412 -0.67350476
        # [16,] -0.11101712 -0.3138824 -0.48491557
        # [17,] -0.11101712  1.4124708  0.22077297
        # [18,]  0.01644698  1.5694121 -0.16857243
        # [19,]  0.14391108 -0.7847060 -0.27199230
        # [20,]  0.17577710 -1.8832945 -0.32066048
        # [21,]  0.23950915  0.3138824  0.26335763
        # [22,]  0.30324119  0.6277648  0.09301901
        # [23,]  0.39883927 -0.3138824  0.37286102
        # [24,]  0.87682962 -0.6277648  0.49453146
        # [25,]  0.97242770  0.1569412  0.75612291
        # [26,]  1.29108793  0.7847060  1.53481372
        # [27,]  1.35481998  0.9416472  1.55306429
        # [28,]  1.48228408  0.6277648  1.71123586
        # [29,]  1.51415010  0.6277648  1.29755636
        # [30,]  1.51415010  0.6277648  1.26713875
        # [31,]  2.34266672  1.7263533  2.84885447

我意外的错误结果：

    # [1] -1.57685421 -1.48125614 -1.41752409 -0.87580169 -0.81206964 -0.78020362
    # [7] -0.71647157 -0.71647157 -0.68460554 -0.65273952 -0.62087350 -0.58900747
    # [13] -0.58900747 -0.49340940 -0.39781133 -0.11101712 -0.11101712  0.01644698
    # [19]  0.14391108  0.17577710  0.23950915  0.30324119  0.39883927  0.87682962
    # [25]  0.97242770  1.29108793  1.35481998  1.48228408  1.51415010  1.51415010
    # [31]  2.34266672 -0.94164723 -1.72635326 -2.04023567 -0.62776482  0.78470603
    # [37]  1.09858844 -1.56941205 -0.15694121  0.62776482 -0.15694121  0.47082362
    # [43]  0.00000000  0.00000000 -1.09858844 -0.15694121 -0.31388241  1.41247085
    # [49]  1.56941205 -0.78470603 -1.88329446  0.31388241  0.62776482 -0.31388241
    # [55] -0.62776482  0.15694121  0.78470603  0.94164723  0.62776482  0.62776482
    # [61]  0.62776482  1.72635326 -1.20885469 -1.20885469 -1.21493821 -0.83775985
    # [67] -0.69175532 -0.63700362 -0.88642802 -0.72825645 -0.46058149 -0.62483658
    # [73] -0.36324513 -0.55791784 -0.53358375 -0.53966727 -0.67350476 -0.48491557
    # [79]  0.22077297 -0.16857243 -0.27199230 -0.32066048  0.26335763  0.09301901
    # [85]  0.37286102  0.49453146  0.75612291  1.53481372  1.55306429  1.71123586
    # [91]  1.29755636  1.26713875  2.84885447
    # attr(,"scaled:center")
    # Girth   Height   Volume 
    # 13.24839 76.00000 30.17097 
    # attr(,"scaled:scale")
    #     Girth    Height    Volume 
    # 3.138139  6.371813 16.437846

为什么函数中的命令attributes(result)会给出错误的结果？

有人可以解释吗？

Answer 1

由于我们要覆盖现有属性，因此需要将新属性附加到现有属性中，请参见：

z_function = function(x){
  (x - mean(x))/sd(x)}

scale_function = function(x){
  result = apply(x,2,z_function)
  att_mean = apply(x,2,mean)
  att_sd = apply(x,2,sd)
  # using c() to combine 2 lists of attributes
  attributes(result) = c(attributes(result),
                         list("scaled:center" = att_mean,"scaled:scale"= att_sd))
  result
}

# check if output is the same with "scale"
identical(scale(trees), scale_function(trees))
# [1] TRUE

或者按照@shadow的注释中的建议，我们可以使用attr，它附加属性而不是覆盖：

attr(result, "scaled:center") <- att_mean
attr(result, "scaled:scale") <- att_sd

将属性应用于数据框，重写比例函数

1 个答案: