Add a variable to a data frame containing max value of each row

Using tidyverse you can try the following:

considering all numeric columns

df %>%
  keep(is.numeric) %>% 
  rowwise() %>%
  mutate(maxval = max(across()))

in your specific case

df %>%
  rowwise() %>%
  mutate(maxval = max(across(2:26)))

Note: For tons of rows, the rowwise() operation will slow down your analysis.

Vectorized version with pmax:

df$max <- do.call(pmax, df[2:26])

In case when you need omit NA values syntax is:

do.call(pmax, c(df[2:26], list(na.rm=TRUE)))

The second argument of do.call need to be a list of arguments to function. df is already list so we concatenate it with na.rm=TRUE argument (converted to list).

You can use apply. For instance:

df[, "max"] <- apply(df[, 2:26], 1, max)

Here's a basic example:

> df <- data.frame(a=1:50, b=rnorm(50), c=rpois(50, 10))
> df$max <- apply(df, 1, max)
> head(df, 2)
  a          b  c max
1 1  1.3527115  9   9
2 2 -0.6469987 20  20
> tail(df, 2)
    a          b  c max
49 49 -1.4796887 10  49
50 50  0.1600679 13  50

Here are two additional methods. The first,in base R, is to combine matrix extraction [ with max.col, which returns a vector indexing the column position of the maximum value in each row.

df$max <- df[2:26][cbind(seq_len(nrow(df)), max.col(df[2:26]))]

cbind constructs a matrix indexing the position of the maximum value for each row and [ uses this to extract this value.

The second is to use rowMaxs in the matrixStats package. This looks like

library(matrixStats)
rowMaxs(as.matrix(df[2:26])))

---

Let's do some benchmarking.

# data.frame with 1000 observations and 26 variables
set.seed(1234)
df <- data.frame(id=paste0(letters[-1], 1:40), matrix(rnorm(25000L, 5L, 10L), 1000L))

Also add the rowMaxs function from the matrixStats package to the mix.

library(matrixStats)
library(microbenchmark)

microbenchmark(apply=apply(df[, 2:26], 1, max),
               pmax=do.call(pmax, df[2:26]),
               max.colSub=df[2:26][cbind(seq_len(nrow(df)), max.col(df[2:26]))],
               rowMaxs=rowMaxs(as.matrix(df[2:26])))
Unit: microseconds
        expr      min        lq      mean    median        uq      max neval cld
       apply 1610.540 1786.5905 2193.5334 1863.5680 1990.4380 6915.999   100   c
        pmax  354.382  364.6455  380.1720  373.3405  385.4580  567.923   100 a  
  max.colSub  604.416  651.7430  822.6015  664.7155  681.2510 3086.512   100  b 
     rowMaxs  243.762  264.0040  320.2350  277.9750  290.5190 2328.712   100 a

So, rowMaxs is the clear winner followed by pmax and then by max.col, with matrix extraction, and apply at the tail end of the pack.

With a data.frame with 10000 rows and 26 columns, we get a similar story:

set.seed(1234)
df <- data.frame(id=paste0(letters[-1], 1:400), matrix(rnorm(250000L, 5L, 10L), 10000L))

The above code returns

Unit: milliseconds
       expr       min        lq      mean    median        uq      max neval cld
      apply 15.193361 18.299830 21.737516 20.337880 21.774793 99.44836   100   c
       pmax  3.060853  3.101481  3.156630  3.137545  3.191430  3.54182   100 a  
 max.colSub  3.338828  3.642603  7.051700  3.992708  6.336531 84.43119   100  b 
    rowMaxs  1.244184  1.322302  2.675281  1.508474  1.638053 79.28054   100 a