Skip to contents

Thank you for your interest in contributing to this repo! Your contribution is highly valued. Please go through this document for guidance on how to contribute.

Programming conventions

Please follow the programming conventions to ensure a consistent programming style across the package.

Generally we follow the tidyverse style guide. Some specific conventions that deviate from this are explained below.


  1. Function names should be explicit and clear. Words should be separated by underscore (snake_case).
  2. Functions starting with h_ are helper functions and they should not be exported.
  3. Functions should be well documented using roxygen2 (even when they are not exported).
  4. Functions must be unit tested (even when they are not exported).


In package mmrm, we follow the following convention in package imports.

  1. When only a few functions from another package are needed:
    1. In DESCRIPTION, add that package into Imports.
    2. In mmrm-package.R, we add a importFrom with a single function from the package.
      1. The function imported can be any random function within the package.
    3. Always use package::function style wherever you need to use the function.
    4. Example: stats.
  2. When many functions from another package are needed:
    1. In DESCRIPTION, add that package into Imports.
    2. In mmrm-package.R, we use a import to import every function.
    3. Use the functions directly without prefix.
    4. Example: checkmate.


  • Comments should always follow sentence style.
  • Comments should be as minimal as possible: Generally the code and variable names should be clear enough to not need any comments. Only use them when needed and explain the “Why” and not the “What”.


  • All functions must be documented using roxygen2 chunks, including internal functions (see also above).

  • Exported objects must have a lifecycle badge to clarify the maturity.

    • Start with "experimental" status and consider upgrading to "stable" once the interface has been stable for several months.
  • Use Title Style for the title of the documentation.

  • Always include a @description part with at least one sentence describing the object.

  • For the arguments use the following convention:

    @param formula (`formula`)\cr the model formula, see details.

    So the type of the argument is in parentheses, followed by line break, followed by lower case half-sentence ending with a full stop.

  • For references to other help pages use the corresponding markdown syntax, e.g. [function()] to reference other functions.

  • Exported objects must be included in the _pkgdown.yml file to be populated on the pkgdown website.

  • In vignettes, you cannot directly reference help pages but only pkgdown web pages. Note that this includes only exported objects. To make it look similar to the help page references, please here also use function() style.

  • mmrm_review_methods.Rmd is a large vignette and we precompute this vignette to make the GitHub actions faster. Run the script vignettes/precompile.R to regenerate the precomputed vignette and subsequently update it in GitHub. Before every release we need to run this again. Please note you need to install the package and then compile the vignette because efficiency is better after installation to provide a fair comparison.

Github conventions

When using GitHub to collaborate, the following conventions are needed:

  1. Github issues is for issues, feature requests, bugs. Before creating a issue, please make sure this issue has not been reported yet.
    • If you are going to work on this issue, please assign yourself.
  2. Please create a branch in the mmrm repository, instead of creating forks, unless you are not yet a team member.
    • Branches should be associated with a GitHub issue and linked to an issue id.
    • The name of a branch should be of the form: <issue_id>_<short_discription>.
  3. Add changes to the branch and push it to GitHub.
    • Please use clear commit messages.
    • Please keep your changes focused on the issue. If there are independent changes, please separate it into another PR linking to another issue.
  4. Please create a Pull Request when you think your code changes are ready:
    • Functions are well documented.
    • Functions have corresponding unit tests.
    • Changes pass all the GitHub action checks.
    • The checklist in the corresponding issue is completed.
  5. Address all the comments you receive.
    • at least one approval is needed to merge.

Contribution tips

Development environment Set-up

The development this mmrm package is based on the latest R version and C++ compilers. The package dependencies are the most recent versions from CRAN. We recommend that your working environment is set up in the same way. Additionally, there are some tools we recommend you to install:

  1. RTools if you work on a Windows operating system. Alternatively you can use docker to separate the operating system and the development system.
  2. GitKraken is a very useful user interface for git including visualization of git commit graphs, file history, etc.
  3. lintr will allow you to perform static code analysis.
  4. pre-commit is a Python module that allows you to identify issues before you commit locally.

Issue labels

The issues are categorized with several labels:

Label name Description
SPx SP (story points) indicate complexity, and the larger the subsequent number, the more time consuming the issue is expected to be
priority Issues with this label should be completed with higher priority
good first issue Good choices for new team members
blocked Blocked by other issues
bug Something isn’t working
devops Development and Operation
discussion Discussion needed
documentation Improvement of documentation is needed
duplicate The issue already exists
enhancement New feature or request
help wanted Extra attention is needed
invalid This doesn’t seem right
question Further information needed

Please choose an issue based on your interest, issue complexity, and priority.

Add new unit tests

To add a new unit test, you need to first identify the test scope. Does the test fit in the scope of existing tests? If so, please modify the existing test files under tests/testthat/ folder or src/ folder, depending on whether the code to be tested is R or C++. Otherwise please create a new test file, with a name prefix of “test-”.

In each test case, use the following structure:

test_that("function_name does something as expected", {
  result <- function_name(input)
  expected <- hardcoded_result
  expect_identical(result, expected)

The purpose of the test should be clearly stated first.

In the test body part, conduct the tests, e.g. use expect_identical to check consistency, expect_error to catch error messages, etc. The test body should not follow the same implementation logic as the package did, otherwise we may miss mistakes in implementation.

Add integration tests

Integration tests compare the results of SAS and R and assures the quality of our code. To add an integration test, you need to do the following:

  1. Use SAS to run an appropriate mmrm model with proc mixed, using fev_data.
  2. Save the results in .txt format in the design/SAS/ folder.
  3. Decide the key outputs that are needed for comparison.
  4. Add a unit test verifying that the R implementation of the same model has the same results (conversion may be needed).

Get started with C++

If you have no experience with C++, it is totally fine: TMB has provided us with many high-level functionalities that is very similar to R. Here we only list the most important things that you need to go through before you begin C++ programming.

  1. Semicolons. C++ use semicolons to terminate a statement. In R, we can use semicolons or line breaks to do so, but in C++, we need both semicolons and line breaks.
  2. Types. C++ is a strong type programming language and all objects need to be declared with type. Sometimes you can combine the declaration and definition.
    1. int i = 1; This works, as we declare i as int and define it to be 1.
    2. i = 1; This fails, because i is not declared yet.
    3. int i; i = 1; This works, because i is declared and then defined.
    4. int i = 1, j = 2; This works, because i and j are both int.
  3. Object scope. In R, objects declared inside control flows will exist outside of that enclosing brackets of the flow, while in C++, objects created within control flows will be terminated.
    1. in R, if (TRUE) { a = '123' }; print(a) is legal.
    2. in C++, if (1) {string a = '123'}; std::stdout << a << std::endl; is illegal, because object a is terminated already.
    3. in C++, string a; if (1) {a = '123'}; std::stdout << a << std::endl; is legal, a is declared prior to if statement.
  4. Polymorphism. Unlike in R, where function arguments have no explicitly defined types, in C++ the type of each argument of a function must be pre-specified.
    1. Template is a special function that works with generic argument type. We could imagine a single function that could work on arguments of arbitrary type, and template functions make this possible through separation of function logic from the argument declaration. In this way we can use template functions and avoid the need to replicate the whole code for each type.

With these points in mind, you are about ready to go.

Get started with TMB

In mmrm we are not including any latent variables and so the Laplace approximation aspect of TMB is not used. We only use the automatic differentiation part of TMB. For detailed documentation of TMB, visit the TMB reference.

One important feature of TMB are the R style matrix/array calculations. This is important because we mainly use this part to conduct our calculations. See matrix_arrays.cpp for examples.

Add a new covariance structure

To add a new covariance structure, you need to do the following:

  1. Understand the covariance structure and add appropriate documentation in covariance structure. Create a draft pull request to invite discussion from other team members.
  2. Implement the covariance structure on the C++ side and the corresponding R interface.
  3. Add unit tests to make sure the new covariance structure is working as expected.
  4. Add integration tests under design/SAS/ folder to make sure SAS and R produce similar results (within tolerance).

Add additional data

To add additional data to mmrm, please follow the next steps:

  1. Make sure the data is needed.
  2. Document the data in R/data.R.
  3. Use save to create an rda data only containing this dataset. Here the function usethis::use_data can also be helpful.

Communications within team

There are several communication channels, please use appropriate ones.


GitHub issues and pull requests are where implementations are discussed and reviewed. Feature requests, bugs, enhancements, technical implementations can be discussed here. When you have ideas that needs to be documented, it is better to have them in GitHub.


Slack is a messaging tool and we have the mmrm channel under the rinpharma space. You can put anything in the slack channel, e.g., you have completed a issue and are waiting for review, or you have some questions and don’t want to wait until the next stand-up meeting.

To join the slack channel, please make sure you have a slack account, and send the email address to any team member.

Stand-up meetings

We have stand-up meetings currently three times a week. The purpose of stand-up meetings is for team members to discuss the current work progress and raise any issue/questions for team discussion if needed.

You don’t need to join all three meetings.