Question

我们如何计算R中的平均精度得分？有一个简单的方法吗？

我按如下方式计算。我不知道它是否完全正确..

pr = prediction(preds, labs)
pf = performance(pr, "prec", "rec")
# plot(pf)

pf@x.name
 [1] "Recall"

pf@y.name
 [1] "Precision"

rec = pf@x.values[[1]]

prec = pf@y.values[[1]]

idxall = NULL
for(i in 1:10){
  i = i/10

  # find closest values in recall to the values 0, 0.1, 0.2, ... ,1.0
  idx = which(abs(rec-i)==min(abs(rec-i)))

  # there are more than one value return, choose the value in the middle
  idx = idx[ceiling(length(idx)/2)] 

  idxall = c(idxall, idx)
}

prec.mean = mean(prec[idxall])

Answer 1

我添加一个例子。此示例假设您将实际Y值作为二进制值的向量，并将预测Y值作为连续值的向量。

# vbYreal: real Y values
# vdYhat: predicted Y values
# ex) uNumToExamineK <- length(vbYreal)
#     vbYreal <- c(1,0,1,0,0,1,0,0,1,1,0,0,0,0,0)
#     vdYhat <- c(.91, .89, .88, .85, .71, .70, .6, .53, .5, .4, .3, .3, .3, .3, .1)
# description:
# vbYreal_sort_d is the descending order of vbYreal(e.g.,     c(1,0,1,0,0,1,0,0,1,1,0,0,0,0,0) )
FuAPk <- function (uNumToExamineK, vbYreal, vdYhat){

  # The real Y values is sorted by predicted Y values in decending order(decreasing=TRUE) 
  vbYreal_sort_d <- vbYreal[order(vdYhat, decreasing=TRUE)]
  vbYreal_sort_d <- vbYreal_sort_d[1:uNumToExamineK]
  uAveragePrecision <- sum(cumsum(vbYreal_sort_d) * vbYreal_sort_d / seq_along(vbYreal_sort_d)) /
    sum(vbYreal_sort_d)
  uAveragePrecision
}

vbYreal <- c(1,0,1,0,0,1,0,0,1,1,0,0,0,0,0)
vdYhat <- c(.91, .89, .88, .85, .71, .70, .6, .53, .5, .4, .3, .3, .3, .3, .1)

FuAPk(length(vbYreal), vbYreal, vdYhat)
# [1] 0.6222222

Answer 2

Here是Metrics包中的一个示例。

R的平均精度

2 个答案: