Comp. Efficent way of resetting sequence if condition met ( R )

Question

Problem:

I want to reset a (1,2) sequence if condition is met (subject changes).
I have for and if loops that will do this but, unsurprisingly, that method is very slow. Any suggestions (e.g., involving the apply family) for a more efficent method?

Current:

  subj odd_even
    a         
    a         
    a         
    b         
    b         
    b         
    b         
    c         
    c         
    c         

---

Goal:

  subj odd_even
    a      1   
    a      2   
    a      1   
    b      1   
    b      2   
    b      1   
    b      2   
    c      1   
    c      2   
    c      1   

---

df = data.frame( subj = c("a","a","a","b","b","b","b", "c","c","c"), odd_even = "" )

Answer 1

I like the sequence function for this:

df$odd_even <- 2L - sequence(table(df$subj)) %% 2L

data.table is another option:

library(data.table)
setDT(df)
df[, odd_evenDT := 2L - seq_along(.I) %% 2L, by = subj]

Benchmarks:

set.seed(42)
df <- data.frame(subj = sort(sample(as.character(1:1e4), 1e5, TRUE)))
DT <- data.table(df)

library(microbenchmark)
microbenchmark(roland1 = 2L - sequence(table(df$subj)) %% 2L,
           roland2 = DT[,2L - seq_along(.I) %% 2L, by = subj],
           roland3 = 2L - sequence(rle(as.integer(df$subj))$lengths) %% 2L,
           jeremy = df %>% group_by(subj) %>%
             mutate(odd_even = 2 - (row_number() %% 2)),
           frank = 2L - ave(as.integer(df$s),df$s,FUN=seq_along) %% 2L, 
           flick = ave(seq_along(df$subj), df$subj, FUN=function(x) rep(c(1,2), length.out=length(x))),
           times = 10, unit = "relative")

# Unit: relative
#     expr      min       lq      mean   median        uq      max neval
#  roland1 5.820459 5.754497 5.0368686 5.404110 4.0853039 4.847161    10
#  roland2 1.110919 1.057952 0.9840653 1.037428 0.7939004 1.176258    10
#  roland3 1.000000 1.000000 1.0000000 1.000000 1.0000000 1.000000    10
#   jeremy 5.024087 4.941366 4.3491117 4.635534 3.5144515 4.277011    10
#    frank 2.036816 1.944603 1.7809168 1.831937 1.6459597 1.607283    10
#    flick 3.655127 3.621457 3.2453089 3.473188 2.7717947 3.198285    10

Answer 2

Here's another clunky approach:

df$odd_even <- 2L - ave(as.integer(df$s),df$s,FUN=seq_along) %% 2L

The ave makes a counter within each group. That counter is what we're odd-vs-even testing.

Answer 3

What is the desired behavior if a subj reoccurs later in the dataframe?

If it won't happen, here's a dplyr method:

library(dplyr)

df %>% group_by(subj) %>%
       mutate(odd_even = 2 - (row_number() %% 2))