I have a dataframe of cases that score 0-1 on a group of binary attributes.
What I want to do is extract all possible ombinations of attribute triplets (e.g. A/B/C, A/B/D... out of A-E) and then sum for each possible combination triplet the number of times a case in the original dataframe matched those attributes.
Using dplyr logic as well as lapply I can solve this problem but the performance is very bad, especially for bigger dataframes and more possible attributes. My real dataframe leads to a test matrix of >1000 possible triplets and the function performs very bad on this.
Please help me optimize the code while ideally staying within the dplyr framework as much as possible.
library(tidyverse)
# Create a test data frame and vector of relevant variables
test_df <- data.frame(ID = c(1,2,3,4), Target = c(1,1,0,0),F_A = c(1,0,0,1),F_B = c(0,1,0,1),F_C = c(1,1,0,0),F_D = c(0,1,1,0),F_E = c(1,0,0,1))
invars = c("F_A","F_B","F_C","F_D","F_E")
NumOfElements = 3
# Create a full matrix of all relevant variables in NumOfElements-combinations
combn(invars,NumOfElements) %>%
t() %>%
as.data.frame() %>%
rowid_to_column("ID") %>%
select(ID, T1 = V1, T2 = V2, T3 = V3) %>%
unite("Test",starts_with("T"),sep = "|",remove = FALSE,na.rm = TRUE) %>%
{.} -> test_matrix
# Brute Force Function to calculate number of all IDs that fullfill the test rules
bruteForce_size = function(rule_iterator,source_df,invars){
source_df %>%
pivot_longer(cols = c(-ID,-Target), names_to = "Affinity", values_to = "Value") %>%
mutate(Value = ifelse(Value ==1, Affinity,NA_character_)) %>%
pivot_wider(names_from = Affinity, values_from = Value) %>%
unite("Test",invars,sep = "|",remove = FALSE,na.rm = TRUE) %>%
mutate(Size = as.numeric(rule_iterator == Test)) %$%
sum(Size)
}
# Calculate and attach sizes to test_matrix
test_matrix %>%
mutate(Size = unlist(lapply(Test, bruteForce_size, test_df)))
