How do I conduct a row sums for loop across specific columns using [row,col] distance indexing

Question

Re-purposing a previous question but hopefully with more clarity + dput().

I'm working with the data below that almost similar to a key:value pairing such that every "type" variable has a corresponding variable that contains a "total" value across each row.

  structure(list(type3a1 = c(2L, 6L, 5L, NA, 1L, 3L, NA), type3b1 = c(NA, 
3L, 1L, 5L, 6L, 3L, NA), type3a1_arc = c(1L, 2L, 5L, 4L, 5L, 
4L, NA), type3b1_arc = c(2L, 2L, 3L, 4L, 1L, 1L, NA), testing = c("Yes", 
NA, "No", "No", NA, "Yes", NA), cars = c(5L, 12L, 1L, 6L, NA, 
2L, NA), house = c(5L, 4L, 0L, 5L, 0L, 10L, NA), type3a2 = c(50L, 
NA, 20L, 4L, 5L, NA, NA), type3b2 = c(10L, 10L, 15L, 1L, 3L, 
1L, NA), type3a2_arc = c(50L, 25L, 30L, 10L, NA, 10L, NA), type3b2_arc = c(NA, 
20L, 10L, 50L, 5L, 1L, NA), X = c(NA, NA, NA, NA, NA, NA, NA)), class = "data.frame", row.names = c(NA, 
-7L))

I am trying to do a summation loop that goes through every row, and scans each "type" variable (ie type3a1, type3b1, type3c1, etc.). Each "type" has a matching variable that contains its "total" value (ie type3a2, type3b2, type3c2, etc.)

Process:

1. Check if the "type" variable contains values in (1,2,3,4 or 5).
2. If that type column's [row,col] value is in (1:5), then move 7 columns from its current [row,col] index to grab it's total value and ready for summation.
3. After checking every "type" variable, sum all the gathered "total" values and plop into a new overall totals column.

Essentially, I want to end up with a total value like the one below:

The first row shows a total of 100 since type3b1 has a value of "NA" which is not in (1:5). Hence, its total pairing (ie +7 columns away = cell value of "10") is not accounted for in the row summation.

My approach this time compared to a previous attempt is using a for loop and relying on indexing based on how far a column was away from another column. I was having a lot of trouble approaching this using dplyr / mutate methods and was having a lot of issues with the variability in the type:total name pairings (ie no pattern in the naming conventions, very messy data)...

# Matching pairing variables (i.e. type_vars:"type3a1" with total_vars:"type3a2") 
type_vars <- c("type3a1", "type3b1", "type3a1_arc", "type3b1_arc")
total_vars <- c("type3a2", "type3b2", "type3a2_arc", "type3b2_arc")

valid_list <- c(1,2,3,4,5)
totals = list()

for(row in 1:nrow(df)) {
  sum = 0
  for(col in type_vars) {
    if (df[row,col] %in% valid_list) {
      sum <- sum + (df[row,col+7])
    }
  }
  totals <- sum
}

I'm hoping this is the right approach but in either case, the code gives me an error at the sum <- sum + (df[row,col+7]) line where: Error in col + 7: non-numeric argument to binary operator .

It's weird since if I were to do this manually and just indicate df[1,1+2] , it gives me a value of "1" which is the value of the intersect [row1, type3a1_arc] in the df above.

Any help or assistance would be appreciated.

Answer 1

Here is one way with tidyverse - loop across the columns with names that matches the 'type' followed by one or more digits ( \\d+ ), a letter ( [az] ) and the number 2, then get the corresponding column name by replacing the column name ( cur_column() ) substring digit 2 with 1, get the value using cur_data() , create a logical vector with %in% , negate ( ! ) and replace those not in 1:5 to NA, then wrap with rowSums and na.rm = TRUE to get the total

library(dplyr)
library(stringr)
df1 %>% 
 mutate(total = rowSums(across(matches('^type\\d+[a-z]2'), ~ 
  replace(.x, !cur_data()[[str_replace(cur_column(),
   "(\\d+[a-z])\\d+", "\\11")]] %in% 1:5, NA)), na.rm = TRUE))

-output

   type3a1 type3b1 type3a1_arc type3b1_arc testing cars house type3a2 type3b2 type3a2_arc type3b2_arc  X total
1       2      NA           1           2     Yes    5     5      50      10          50          NA NA   100
2       6       3           2           2    <NA>   12     4      NA      10          25          20 NA    55
3       5       1           5           3      No    1     0      20      15          30          10 NA    75
4      NA       5           4           4      No    6     5       4       1          10          50 NA    61
5       1       6           5           1    <NA>   NA     0       5       3          NA           5 NA    10
6       3       3           4           1     Yes    2    10      NA       1          10           1 NA    12
7      NA      NA          NA          NA    <NA>   NA    NA      NA      NA          NA          NA NA     0

---

Or may also use two across (assuming the columns are in order)

df1 %>%
   mutate(total = rowSums(replace(across(8:11), 
     !across(1:4, ~ .x %in% 1:5), NA), na.rm = TRUE))

-output

   type3a1 type3b1 type3a1_arc type3b1_arc testing cars house type3a2 type3b2 type3a2_arc type3b2_arc  X total
1       2      NA           1           2     Yes    5     5      50      10          50          NA NA   100
2       6       3           2           2    <NA>   12     4      NA      10          25          20 NA    55
3       5       1           5           3      No    1     0      20      15          30          10 NA    75
4      NA       5           4           4      No    6     5       4       1          10          50 NA    61
5       1       6           5           1    <NA>   NA     0       5       3          NA           5 NA    10
6       3       3           4           1     Yes    2    10      NA       1          10           1 NA    12
7      NA      NA          NA          NA    <NA>   NA    NA      NA      NA          NA          NA NA     0

---

Or using base R

df1$total <- rowSums(mapply(\(x, y) replace(y, !x %in% 1:5, NA), 
      df1[1:4], df1[8:11]), na.rm = TRUE)
df1$total
[1] 100  55  75  61  10  12   0

Answer 2

Here's a base R solution:

valid_vals <- sapply(type_vars, \(col) df[, col] %in% valid_list)

temp <- df[, total_vars]
temp[!valid_vals] <- NA

df$total <- rowSums(temp, na.rm = TRUE)
df$total
# [1] 100  55  75  61  10  12   0

Answer 3

The error you received is because col in your original for loop iterates through type_vars which is a character data type . One way around this is to reference column indices of type_vars using the which() function. Here is a solution with just a couple of modifications to your for loop:

totals <- c()

for(row in 1:nrow(df)) {
    sum = 0
    for(col in which(names(df) %in% type_vars)) {
        if (df[row,col] %in% valid_list) {
            sum <- sum(c(sum, (df[row,col+7])), na.rm=T)
        }
    }
    totals[row] <- sum
}
df$totals <- totals
df$totals
[1] 100  55  75  61  10  12   0