如何在R中的两列之间查找精度

Question

I have two columns: one for predicted value and another for true value.
I want to calculate the accuracy between these columns and make missing values count as true if both columns have missing values.
So when

Pred True
1     2
2     2
NA    NA
3     2

The accuracy would be 50%.
Also, how should I do the same thing with character values?

Answer 1

You can do:

pred <- c(1,2,NA,3)
true <- c(2,2,NA,2)
(sum(pred==true, na.rm=T) + sum(is.na(pred) & is.na(true))) / length(pred)

That is, add the number of times where pred and true are equal sum(pred==true, na.rm=T) together with the number of times they are both NA values sum(is.na(pred) & is.na(true)) . Divide by the vector length.

Answer 2

You could do something like this:

sum(data$Pred == data$True, na.rm = T) / nrow(data) *100

to get accuracy the way you defined it. It will work for integers and strings. Problem is, you can't really compare NAs, so if both columns have NAs for a given row and you actually consider that to be an accurate prediction, you would need to count those instances separately. For instance, you can just get the union of the indices where both columns are NA, and add that to the sum:

s <- sum(data$Pred == data$True, na.rm = T)
na <- length(union(which(is.na(data$Pred)), which(is.na(data$True))))
(s + na) / nrow(data) * 100