Here a link to the lecture slides for this session: LINK


In this practical you’ll learn how to produce tidy code (and data). By the end of this practical you will know how to:

  1. Write clean, documented code.
  2. Understand errors and warnings.
  3. Deal with missing values.

The Do’s and Don’ts of clean code

Filenames should be meaningful. To order them, prefix them with numbers.

# Good

# Bad

Object names should be lowercase. Use _ rather than . or ‘CamelCase’ (using capitalization) for multi-word names. If possible use nouns for variables and verbs for functions. Use meaningful names. Avoid using names of existing objects.

# Good

# Bad

Place spaces around all operators, such as, =, +, -, <-, etc. Also applies for defining arguments in functions. Always put a space after, never before a comma.

# Good
var_rt <- var(rt, na.rm = TRUE)

# Bad

Extra spacing may be used to align assignments.

# Good
  var_rt  = var(rt)
  mean_rt = mean(rt)

An opening curly bracket should never be on its own line. Always indent within curly brackets. To indent code use two spaces. Don’t use tabs.

# Good
if (my_dbl < 2){
  message('my_dbl is smaller 2')
} else {
  message('my_dbl is larger or equal 2')

# Bad
if (my_dbl < 2)
message('my_dbl is smaller 2')
} else {
message('my_dbl is smaller 2')  

For assignments use <-, not =.

# Good
x <- 24324

# Bad
x = 24324

Comment each line of your code. To break up your code in chunks use - or =.

# Plot data ----------------------------

# Plot data ============================

The most frequent errors

R’s error messages are not always very intuitive, but over time you will learn to understand them. In the beginning it helps to focus on the part after the colon. E.g.,

sapply(1:10, 'fefkl')
Error in get(as.character(FUN), mode = "function", envir = envir): object 'fefkl' of mode 'function' was not found

Here the key message is 'function' was not found. I.e., R is interpreting 'fekl' as a function but cannot find an instance of this function anywhere (because it doesn’t exist).

According to an analysis of, a popular help forum, the 7 most frequent error messages and their meaning are:

Error Example Description
'could not find function' lenth(my_vec) There is a typo in the function name or that a package has not been loaded.
'error in if' if(NA == 2) 2 + 2 The object in the if clause is non-logical or NA.
'error in eval' lm(fefq~wzfe) An object is used that does not exist.
'cannot open()' read_csv(‘hjht.txt’) The file does not exist. Could be a typo or a missing filepath.
'no applicable method' predict(‘efwe’) A ‘generic function’ has not been defined for this type/class
'subsscript out of bounds' a <- matrix(c(1,2)); a[2,2] R tried to access an element (or variable) that does not exist
package errors Occur when R is unable to install, compile, or load a package. Often this means that some software in the background is missing.

For more information visit here.

A mice example

To impute missing values, the mice package is very helpful. The code below loads the titanic dataset containing records on 1313 Titanic survivors and then attempts to predict missing values in Age using central tendencies and mice.

# Load packages

# read and duplicate in titanic data
data <- read_csv('')
data_mean <- data
data_median <- data
data_mice <- data

# use central tendencies
data_mean$Age[$Age)] <- mean(data_mean$Age, na.rm = TRUE)
data_median$Age[$Age)] <- median(data_median$Age, na.rm = TRUE)

# use mice
mice_model <- mice(data_mice, method = 'rf') # uses random forests
data_mice  <- complete(mice_model)


Begin new project

If you haven’t already, begin a new project in a new folder. Within the folder, create two new folders called 1_data and 2_code.

Clean code

  1. Below you see some ‘dirty’ code. Go through it and clean it according to the above principles (incl. commenting). Then, write the data file that the code reads in and write it to your 1_data folder using write_csv(). Finally, change in the code the URL its loading the data from to the new file path on your computer. When ready save the code in the 2_code folder as cleaned_code_2018Jan27.R.

Note: The code is using a data set by Sir Francis Galton on the heights of parents and their children, illustrating one of the most classic cases of regression toward the mean.

library(readr   )
library(      magrittr)
library(  dplyr)
  syc =   read_csv('')
  syc = syc%>%mutate(fcm=father/2.54,mcm=mother/2.54)
    a   = syc$father
      b = syc[['mother']]
  1. Now that you have done all the work, I can tell you that R actually has a function to tidy up code. Let’s see what it does. Start again with the ‘dirty’ code and copied into a new script. Save it in your 2_code folder. Now, install and load the formatR package (install.packages("formatR") and library("formatR")) and apply it to the file path of the ‘dirty code’ using the tidy_source(file_path) function. Did it do a good job? What did it miss?

Correct code

  1. OK, now you know how to deal with ‘dirty’ code. Let’s move onto some broken code. Below you see some code that doesn’t work (aside from being pretty ‘dirty’). Do you best to remove the errors (and clean it by the way). When ready, save the code in the 2_code folder as corrected_code_2018Jan27.R.
library(readr   )
syc =   read_csv('')
syc %>% mutate(higher_ratio = height / fther)
vs = syc %>% group_by(father) %>% summarize(a = meaen(height))

Replace NAs

On to different adventures. This section will be about missing values and how one could and should deal with them. Using the very powerful and user-friendly mice-package this will entail some machine learning.

  1. To start off with, run ‘A mice example’ from introductory part above (uses the titanic dataset). Then evaluate the mean and variance (var()) of the variable Age for the imputation by mean and the imputation by random forests (rf) through mice. Do mean and variance differ between the imputation methods? Which imputation method produces the larger variance and why? Remember what Nathaniel said about overfitting and model complexity?

  2. OK, to understand this a better, let’s build our own sandbox. Let’s load the Galton dataset again and use it to introduce our own NAs, so that we can evaluate the relative performance of mean versus mice. To do this, create a copy of the Galton dataset (i.e., assign to new object), and remove in the copy 10% of the values in the variable height. Any easy way to do this, is to use the sample()- function. Provided with the number of rows in the dataset and the number of samples, it will return a vector of indices that can be used to eliminated values (i.e., assign NAs to those indices). Now create additional copies of the Galton copy (with the NAs), one for each method (e.g., Galton_mean and Galton_mice) and impute within those the NAs using the respective methods. To evaluate then the performance of the two methods, we then only need to compare the height variable in the original data with those from the two data sets with imputed values. To summarize the differences use mean squared error mean((original - imputed)^2). What do you expect, which method works better? And, did it?

  3. OK, in the last exercise, we have introduced NAs randomly. Now, let’s see what happens, when NAs are not missing at random. Repeat the analysis of the last exercise, but this time introduce NAs systematically. That is use other variable or combination of other variables to determine the locations of NAs in the height variable. For instance, introduce NAs for father > 70 and the sex == 'M'. Have the results changed? If yes, why?

  4. The mice package offers a variety of different methods. See ?mice. Try out different methods and compare their performance.

Additional reading