Jeremy Playing with R the Language

For my next learning, I decided I need to understand the language “R” a bit better. In addition to our current book, I also found and read parts of R for Beginners. One goal was to be clear about some of the funny syntax that creeps in.

For example, what the heck is df$a?

df is that variable that names a data frame, $a selects column a

df <- data.frame(a = 1:5, second = 2:6, alpha = letters[1:5])
df

So now let’s select a column with the number 1 through 5

df$a

[1] 1 2 3 4 5

or the column named “second”

df$second

[1] 2 3 4 5 6

I like that in R you can operate on vectors. Here, I’m going to double all the numbers in the second column.

df$second <- 2 * df$second
df

You can also either access or assign any table cell with variable[row,col]. Here I am going to change the 8 in the “second” column to be zero.

df[3,2] <- 0
df

I think this should work to change the zero back to 8

df$second[3] <- 8
df

Yes, it did!

A cool thing I learned was that you can put a conditional in the brackets. For example, I could change the cells whose number is less than 10 to 0.

df$second[df$second < 10] <- 0
df

Let’s break that down. The df$second in brackets is a vector of true or false, depending on whether the value is less than 10.

df$second < 10

[1]  TRUE  TRUE  TRUE FALSE FALSE

so basically, we are first figuring out which cells as less than 10, then selecting those to set to 0.

less_than_10 <- df$second < 10
df$second[less_than_10] <- 0
df

let’s make all those zeros be twice the value in column a

df$second[df$second == 0] <- df$a[df$second == 0] * 2
df

By the way, in the San Mateo exercise I did early, I had two problems. It was a poll whose values were either the character “1” (for true) or NA. I really wanted the number 1 or 0 for these values. I’ve now figured out a much simpler way to clean that up.

First, let’s make a data set similar to the San Mateo one. It had two first columns I didn’t care about, plus 4 poll answer columns. The R function “sample” makes a vector by sampling a set. I’ll use that to fake poll answers.

answers <- c("1",NA)
sample(answers,10,replace=TRUE)

 [1] NA  "1" "1" "1" "1" "1" NA  NA  NA  "1"

so that gives me random 1’s or NAs

sm <- data.frame(a = 11:20,
                 b = letters[11:20],
                 answer1 = sample(answers,10,replace=TRUE),
                 answer2 = sample(answers,10,replace=TRUE),
                 answer3 = sample(answers,10,replace=TRUE),
                 answer4 = sample(answers,10,replace=TRUE))
sm

So first let’s convert everything from characters to numbers. as.numeric will do this, and we could do it one column at a time like this.

sm$answer3 <- as.numeric(sm$answer3)

but we could also operate on a subset of the whole table, that is the 3rd through 6th columns.

mutate_at(sm,3:6,as.numeric)

ooh, let’s try that!

sm <- mutate_at(sm,3:6,as.numeric)
sm

Very cool! now my columns with answers are numbers (of type dbl). But they have that pesky NA. Which I want to be a zero for a false answer to my poll.

Note what the following bit does; it produces a true/false table corresponding to whether a cell is an NA.

is.na(sm[3:6])

      answer1 answer2 answer3 answer4
 [1,]   FALSE    TRUE   FALSE    TRUE
 [2,]    TRUE   FALSE    TRUE    TRUE
 [3,]   FALSE   FALSE   FALSE   FALSE
 [4,]   FALSE   FALSE    TRUE   FALSE
 [5,]    TRUE   FALSE   FALSE    TRUE
 [6,]   FALSE    TRUE    TRUE   FALSE
 [7,]    TRUE    TRUE   FALSE    TRUE
 [8,]   FALSE   FALSE   FALSE    TRUE
 [9,]    TRUE   FALSE    TRUE    TRUE
[10,]   FALSE    TRUE    TRUE   FALSE

So I should be able to use that to change the NAs to 0

sm[3:6][is.na(sm[3:6])] <- 0
sm

Wow! That was super cool !! I first got a subset of the data frame with just the columns 3 through 6: sm[3:6]. Then I selected from that those cells who have an na value with a second set of brackets and assigned them to zero. I’m feeling the power.

I also found there is a built-in coalesce function which replaces NAs. And this works too. It takes two arguments, the second being the value to replace the NAs with. But mutate_at only takes a function of one argument. So I made a function in place that takes one arg, and always supplies 0 as the other arg to coalesce.

sm <- mutate_at(sm,3:6,function(x) {coalesce(x,0)})
sm

So now that I understand R better, I could have “cleaned” that San Mateo data with just one piped expression, which converts to numeric and then cleans up the NAs.

sm <- sm %>% mutate_at(3:6,as.numeric) %>% 
        mutate_at(3:6,function(x) {coalesce(x,0)})
sm

sm <- sm %>% mutate_at(3:6,as.numeric) %>% 
        mutate_at(3:6,function(x) {x[is.na(x)] <- 0;x})
sm

That’s all for today!

jeremy