如何使用'by'并选择某些列在data.table中应用自定义函数

Question

我有以下数据表。

#    id  category              sales        Num          share
# 1: 1      a              -0.25174915 -0.2130797 -0.67909764
# 2: 2      a              -0.35569766  0.6014930  0.35201386
# 3: 3      a              -0.31600957  0.4398968 -1.15475814
# 4: 4      b              -0.54113762 -2.3497952  0.64503654
# 5: 5      c              -0.25174915 -0.2130797 -0.67909764
# 6: 6      b              -0.35569766  0.6014930  0.35201386
# 7: 7      c              -0.31600957  0.4398968 -1.15475814
# 8: 8      a              -0.54113762 -2.3497952  0.64503654

当前我正在执行以下操作，以便获得每个类别的PCA分析：

PCA <- prcomp(df[category == a, .(sales, Num, share)], center = T, scale. = T)

df$score <- apply(df[category == a, .(sales, Num, share), ], 1, function(x) sum(x*PCA$rotation))

但是此操作仅对一个特定类别执行，我想对df中的所有类别分别执行。我正在尝试以下操作：

pca.weighting <- function(df,y) {
  PCA <- prcomp(df[, y], center = T, scale. = T)
  scores <- apply(df[, y, ], 1, function(x) sum(x*PCA$rotation))
  return(scores)
}

df[, lapply(.SD, function(x) pca.weighting(df,x) ), by = 'category', .SDcols = c('sales', 'Num', 'share')]

但是我什么都没有，只有错误。我究竟做错了什么 ？任何见识将不胜感激。

Answer 1

也许像：

pca.weighting <- function(df,y) {
    PCA <- prcomp(df, center=TRUE, scale.=TRUE)
    rowSums(as.matrix(df) %*% PCA$rotation)
}

DT[, pca.weighting(.SD, x), category, .SDcols=sales:share]

输出：

   category          V1
1:        a -0.41538278
2:        a  1.15660781
3:        a -0.38428373
4:        a -0.06165818
5:        b -1.26069494
6:        b -0.45360065
7:        c  0.45628037
8:        c  1.54026745

数据：

library(data.table)
DT <- fread("id  category              sales        Num          share
1      a              -0.25174915 -0.2130797 -0.67909764
2      a              -0.35569766  0.6014930  0.35201386
3      a              -0.31600957  0.4398968 -1.15475814
4      b              -0.54113762 -2.3497952  0.64503654
5      c              -0.25174915 -0.2130797 -0.67909764
6      b              -0.35569766  0.6014930  0.35201386
7      c              -0.31600957  0.4398968 -1.15475814
8      a              -0.54113762 -2.3497952  0.64503654")