Ufdjrw

I am trying to create a summary table of accuracy, sensitivity, and specificity using the AUC function within the psych package. I would like to define the input vector (t, a 4 x 1 vector) for each level of the grouped variable.

What I have tried seems to ignore the grouping.

Example:

library(tidyverse)
library(psych)

Data <- data.frame(Class = c("A","B","C","D"),
 TP = c(198,185,221,192),
 FP = c(1,1,6,1),
 FN = c(42,55,19,48),
 TN = c(569,570,564,569))

Data %>% 
 group_by(Class) %>%
 mutate(Accuracy = AUC(t = unlist(.[1,2:5], use.names=FALSE))$Accuracy,
 Sensitivity = AUC(t = unlist(.[1,2:5], use.names=FALSE))$Sensitivity,
 Specificity = AUC(t = unlist(.[1,2:5], use.names=FALSE))$Specificity)

This gives me close to the correct output, except the values for Accuracy, Sensitivity, and Specificity are only being calculated with the first row, then repeated:

# A tibble: 4 x 8
# Groups: Class [4]
 Class TP FP FN TN Accuracy Sensitivity Specificity
 <fct> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl>
1 A 198 1 42 569 0.947 0.995 0.931
2 B 185 0 55 570 0.947 0.995 0.931
3 C 221 6 19 564 0.947 0.995 0.931
4 D 192 1 48 569 0.947 0.995 0.931

I have also tried with summarize:

Data %>% 
 group_by(Class) %>%
 summarize(Accuracy = AUC(t = unlist(.[1,2:5], use.names=FALSE))$Accuracy,
 Sensitivity = AUC(t = unlist(.[1,2:5], use.names=FALSE))$Sensitivity,
 Specificity = AUC(t = unlist(.[1,2:5], use.names=FALSE))$Specificity)

But the output is the same as above.

The desired output is a unique calculation for each level of "Class"

# A tibble: 4 x 8
 Class TP FP FN TN Accuracy Sensitivity Specificity
 <fct> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl>
1 A 198 1 42 569 0.95 0.99 0.93
2 B 185 0 55 570 0.93 0.99 0.91
3 C 221 6 19 564 0.97 0.97 0.97
4 D 192 1 48 569 0.94 0.99 0.92

How do I get the function call within summarize or mutate to maintain the groups?

This works

Data %>% 
 group_by(Class) %>%
 mutate(Accuracy = AUC(t = unlist(.[Class,2:5], use.names=FALSE))$Accuracy,
 Sensitivity = AUC(t = unlist(.[Class,2:5], use.names=FALSE))$Sensitivity,
 Specificity = AUC(t = unlist(.[Class,2:5], use.names=FALSE))$Specificity)

but maybe this is more clear

Data %>% 
 group_by(Class) %>%
 mutate(Accuracy = AUC(t = c(TP, FP, FN, TN))$Accuracy,
 Sensitivity = AUC(t = c(TP, FP, FN, TN))$Sensitivity,
 Specificity = AUC(t = c(TP, FP, FN, TN))$Specificity)

To avoid calling AUC several times for each class, I'd write a wrapper, like this:

# Load libraries
library(tidyverse)
library(psych)

# Create data frame
Data <- data.frame(Class = c("A","B","C","D"),
 TP = c(198,185,221,192),
 FP = c(1,1,6,1),
 FN = c(42,55,19,48),
 TN = c(569,570,564,569))

# Wrapper function
AUC_wrapper <- function(Class, TP, FP, FN, TN)
 res <- AUC(t = c(TP, FP, FN, TN))
 data.frame(Class = Class, 
 TP = TP,
 FP = FP,
 FN = FN,
 TN = TN,
 Accuracy = res$Accuracy, 
 Sensitivity = res$Sensitivity, 
 Specificity = res$Specificity)


# Run using purrr
pmap_dfr(Data, AUC_wrapper)

# Class TP FP FN TN Accuracy Sensitivity Specificity
# 1 A 198 1 42 569 0.9469136 0.9949749 0.9312602
# 2 B 185 1 55 570 0.9309494 0.9946237 0.9120000
# 3 C 221 6 19 564 0.9691358 0.9735683 0.9674099
# 4 D 192 1 48 569 0.9395062 0.9948187 0.9222042