In the last lesson we had an introduction to functions, a way to encapsulate pieces of R code in a way that we substitute different values (arguments) into that code and apply that “recipe” to a set of arbitrary values. However, what if we need the program to “think”, i.e., to make decisions about what to do? What if we need to do something many, many times: ten times, one hundred times, a million times? Yes, our function might be easier to write than the code it runs, but you still don’t want to write your function out that many times. This lesson addresses both of these problems.
If-elseControl structures guide the flow of of your program. The first types of control structure we will consider are branch structures. They are called branch structures because they place a “fork in the road” in your program. The first one to look at is if:
x <- 5
if (x > 3) {
print("x is greater than 3")
}## [1] "x is greater than 3"if (x > 10) {
print("x is greater than 10")
}As you can see in the first case R gets to the word if, checks the condition in the parentheses, and if it is TRUE it runs the code inside the curly brackets. If the expression in the parentheses is FALSE, it skips the code in the brackets. The print() function would print out the message that is provided to it. This isn’t that important at the interactive console, but will be more important as you advance to running more complex programs. You can add an else to provide an alternative block of code to execute if the expression is not TRUE:
if (x > 10) {
print("x is greater than 10")
} else {
print("x is less than or equal to 10")
}## [1] "x is less than or equal to 10"You can chain together a series of if and else statements.
if (x > 10) {
print("x is greater than 10")
} else if (x > 4) {
print("x is greater than 4")
} else {
print("x is less than or equal to 4")
}## [1] "x is greater than 4"You can nest if-else constructions:
if (x > 3) {
if (x > 4) {
print("x is greater than 4")
}
}## [1] "x is greater than 4"SwitchWhat if there are several choices? You can use chained and nested if and else expressions, but often a simpler solution is the switch statement:
center <- function(x, type) {
switch(type,
mean = mean(x),
median = median(x),
trimmed = mean(x, trim = 0.1))
}
x <- c(5, 0, 6, 8, 9, 7, 3, 4, 4, 7)
center(x, "mean")## [1] 5.3center(x, "median")## [1] 5.5center(x, "trim")Like the builtin R functions you can provide defaults within the argument list:
center <- function(x, type = "mean") {
switch(type,
mean = mean(x),
median = median(x),
trimmed = mean(x, trim = 0.1))
}
center(x)## [1] 5.3Within the options you can include more complex code by wrapping it in curly brackets:
center <- function(x, type = "mean") {
switch(type,
mean = mean(x),
median = median(x),
trimmed = mean(x, trim = 0.1),
mode = {
x <- factor(x)
s <- summary(x) # summarize the factor (table of how many each)
m <- max(s) # find the most frequent
l <- levels(x) # extract the levels
o <- l[s == m] # take out the the levels that match the max in the summary
as.numeric(o) # convert back to numbers, factor level names are characters
})
}
center(x, "mode")## [1] 4 7ifelseifelse is a function that provides a simple branch logic for a vector. It is technically not a control structure because it does not change the flow of your program, but it is very useful and similar in concept to if-else.
x## [1] 5 0 6 8 9 7 3 4 4 7ifelse(x == 7, "seven", "not seven")## [1] "not seven" "not seven" "not seven" "not seven" "not seven"
## [6] "seven" "not seven" "not seven" "not seven" "seven"ifelse(x %% 2 == 0, "even", "odd")## [1] "odd" "even" "even" "even" "odd" "odd" "odd" "even" "even" "odd"A new operator worth knowing is the %in% operator. You will find the operator is useful in many situations and can use it in other places than just inside an ifelse function call:
ifelse(x %in% c(7, 4, 0), "in", "out")## [1] "out" "in" "out" "out" "out" "in" "out" "in" "in" "in"forThe for loop control structure allows you to iterate over a vector or list. Here is an example:
for (x in 10:1) {
print(x)
}## [1] 10
## [1] 9
## [1] 8
## [1] 7
## [1] 6
## [1] 5
## [1] 4
## [1] 3
## [1] 2
## [1] 1Read that as for each element in the vector 10:1 (which recall is equivalent to c(10, 9, 8, 7, 6, 5, 4, 3, 2, 1)) call it x temporarily and do what is inside the curly brackets. You could also iterate over the elements of the object directly. For example:
l <- list(a = c(3, 4, 5), b = c("dogs", "cats"))
for (x in l) {
print(x[1])
}## [1] 3
## [1] "dogs"whileThe while loop keeps doing something while (get it?) the condition of the loop is TRUE. For example:
i <- 10
while(i > 0) {
print(i)
i <- i - 1
}## [1] 10
## [1] 9
## [1] 8
## [1] 7
## [1] 6
## [1] 5
## [1] 4
## [1] 3
## [1] 2
## [1] 1If you aren’t careful, you could write the loop wrong in a way that it would never end:
i <- 10
while(i > 0) {
print(i)
i <- i + 1
}This is an infinite loop. To get R to stop, hit the little stop sign at the upper right of your console window (usually the lower left panel of RStudio).
break and nextTwo reserved words permit further control of loops: break and next. When used within a loop (either for or while), they allow you to completely break out of the loop and continue further along your program (break) or skip the rest of the current loop and start the next one (next). They can help with trick situations. Here are a couple of examples:
i <- 10
while(TRUE) {
if(i == 0) break;
print(i)
i <- i - 1
}## [1] 10
## [1] 9
## [1] 8
## [1] 7
## [1] 6
## [1] 5
## [1] 4
## [1] 3
## [1] 2
## [1] 1i <- 20
while(TRUE) {
if(i < 5) break;
if(i %% 2) {
i <- i - 1
next; # go to next iteration if even
}
print(i)
i <- i - 1
}## [1] 20
## [1] 18
## [1] 16
## [1] 14
## [1] 12
## [1] 10
## [1] 8
## [1] 6Now the latter is unnecessarily complex, but demonstrates the concepts. You will plan to make it simpler in “Explore and Extend”.
Sometimes you want to get access to the index of a row or column of a data.frame or the index of an element in a vector or list inside your for loop. The functions NCOL, NROW, length, and seq_along can be a help. Try each with esoph (remember this is a builtin dataset in R). Why is the length of esoph only 5?
Above we created a while loop using both break and next. Rewrite the while loop using neither break nor next.
Recall that the esoph dataset is written in case-control format such that the table is compressed with the first three columns representing patterns of covariates and the last two columns reporting the number of cases and controls with that pattern of covariates. Some statistical methods will accept data in this format, but not all. Sometimes you need to convert the dataset to “long” form where there is one row for each subject and four columns. The first three columns representing the pattern of covariates and the last, which we will call case will be case if the subject was a case and a ctrl if the subject was a control. Our ultimate goal is to convert the esoph data.frame from its current format into the “long” format with those four variables as factors.
Let me give you some additional insight into the problem. Start by examining row 72. Remember how to do that?
esoph[72, ]## agegp alcgp tobgp ncases ncontrols
## 72 65-74 80-119 20-29 2 3We see that this pattern of covariates has 2 cases and 3 controls, so in the “long”" dataset we should have the following 5 rows. (Note: that they neither need to be contiguous nor in any particular order, but we need them somewhere in the new data.frame.)
## agegp alcgp tobgp case
## 1 65-74 80-119 20-29 1
## 2 65-74 80-119 20-29 1
## 3 65-74 80-119 20-29 0
## 4 65-74 80-119 20-29 0
## 5 65-74 80-119 20-29 0Since manipulating factors gets a little tricky, I’d start the program by converting any factor to a character. (You learned how to do that last week.)
You can just use c to append a single item to a vector or list. For lists, if you try to add a more complex item, c does not work and you need a more complex solution (like the one for data.frame below).
l <- esoph[72, ]
is.list(l)## [1] TRUEc(l, case = 1) ## $agegp
## [1] 65-74
## Levels: 25-34 < 35-44 < 45-54 < 55-64 < 65-74 < 75+
##
## $alcgp
## [1] 80-119
## Levels: 0-39g/day < 40-79 < 80-119 < 120+
##
## $tobgp
## [1] 20-29
## Levels: 0-9g/day < 10-19 < 20-29 < 30+
##
## $ncases
## [1] 2
##
## $ncontrols
## [1] 3
##
## $case
## [1] 1To start a data.frame from scratch in this scenario is a little odd. We’ll learn better techniques later, but here is how I would first create the new data.frame you are going to make. stringsAsFactors = FALSE tells R the data.frame function not to convert any character vector into a factor:
new_df <- data.frame(agegp = "", alcgp = "", tobgp = "", case = 0, stringsAsFactors = FALSE)Finally, you need to know “one weird trick” to append a row or column to a data.frame If I have the following data.frame and I want to add a row, I just assign to it as though it was there:
df## agegp alcgp tobgp
## 1 25-34 0-39g/day 0-9g/day
## 2 25-34 0-39g/day 10-19
## 3 25-34 0-39g/day 20-29
## 4 25-34 0-39g/day 30+
## 5 25-34 40-79 0-9g/daydf[6, ] <- list(agegp = "35+", alcgp = "40-79", tobgp = "10-19")
df## agegp alcgp tobgp
## 1 25-34 0-39g/day 0-9g/day
## 2 25-34 0-39g/day 10-19
## 3 25-34 0-39g/day 20-29
## 4 25-34 0-39g/day 30+
## 5 25-34 40-79 0-9g/day
## 6 35+ 40-79 10-19Now, you have all the pieces to write the code to do this. But let’s not get ahead of ourselves. The first step to a problem like this is to carefully think through the process before you even begin to write a line of code. We will keep breaking down the process until we can write code for each step, preferably encapsulating steps in functions where appropriate.
make_new_rows <- function(row) {
ncases <- row$ncases
ncontrols <- row$ncontrols
out <- list()
if (ncases != 0) {
for (i in 1:ncases) {
out[[i]] <- c(row[1:3], case = "case")
}
}
if (ncontrols != 0) {
for (i in (ncases + 1):(ncases + ncontrols)) {
out[[i]] <- c(row[1:3], case = "cntl")
}
}
out
}
# prepare the old data.frame
old_df <- esoph
old_df$agegp <- as.character(old_df$agegp)
old_df$alcgp <- as.character(old_df$alcgp)
old_df$tobgp <- as.character(old_df$tobgp)
# prepare a new data.frame
new_df <- data.frame(agegp = "", alcgp = "", tobgp = "", case = 0, stringsAsFactors = FALSE)
j <- 1 # to hold the row index in the new data.frame
for(i in 1:NROW(old_df)) {
newrows <- make_new_rows(old_df[i, ])
for(r in newrows) {
new_df[j, ] <- r
j <- j + 1
}
}
new_df$agegp <- factor(new_df$agegp)
new_df$alcgp <- factor(new_df$alcgp)
new_df$tobgp <- factor(new_df$tobgp)
new_df$case <- factor(new_df$case)Write a program that converts the “long” format back into the case-control format.
Start here:
new_old_df <- unique(new_df[, 1:3])
new_old_df$agegp <- as.character(new_old_df$agegp)
new_old_df$alcgp <- as.character(new_old_df$alcgp)
new_old_df$tobgp <- as.character(new_old_df$tobgp)
new_old_df$ncases <- 0
new_old_df$ncontrols <- 0Use the following code fragment plus the techniques in this lecture to finish the job. Do not forget that you can assign to these selections like these to replace the value(s). Also, do not forget that anything can be replaced with another name (e.g., “75+”).
new_old_df[new_old_df$agegp == "75+" &
new_old_df$alcgp == "120+" &
new_old_df$tobgp == "10-19", ]## agegp alcgp tobgp ncases ncontrols
## 1174 75+ 120+ 10-19 0 0new_old_df[new_old_df$agegp == "75+" &
new_old_df$alcgp == "120+" &
new_old_df$tobgp == "10-19", "ncases"]## [1] 0new_old_df <- unique(new_df[, 1:3])
new_old_df$agegp <- as.character(new_old_df$agegp)
new_old_df$alcgp <- as.character(new_old_df$alcgp)
new_old_df$tobgp <- as.character(new_old_df$tobgp)
new_old_df$ncases <- 0
new_old_df$ncontrols <- 0
for (i in 1:NROW(new_df)) {
row = new_df[i, ]
if (row$case == "case") {
col <- "ncases"
} else {
col <- "ncontrols"
}
oldval <- new_old_df[new_old_df$agegp == row$agegp &
new_old_df$alcgp == row$alcgp &
new_old_df$tobgp == row$tobgp, col]
new_old_df[new_old_df$agegp == row$agegp &
new_old_df$alcgp == row$alcgp &
new_old_df$tobgp == row$tobgp, col] <- oldval + 1
}
new_old_df$agegp <- factor(new_old_df$agegp)
new_old_df$alcgp <- factor(new_old_df$alcgp)
new_old_df$tobgp <- factor(new_old_df$tobgp)