Developers’ Tips and Tricks#

Productivity and sanity-preserving tips#

In this section we gather some useful advice and tools that may increase yourquality-of-life when reviewing pull requests, running unit tests, and so forth.Some of these tricks consist of userscripts that require a browser extensionsuch asTamperMonkey orGreaseMonkey; to set up userscripts you must haveone of these extensions installed, enabled and running. We provide userscriptsas GitHub gists; to install them, click on the “Raw” button on the gist page.

Folding and unfolding outdated diffs on pull requests#

GitHub hides discussions on PRs when the corresponding lines of code have beenchanged in the meantime. Thisuserscriptprovides a shortcut (Control-Alt-P at the time of writing but look at the codeto be sure) to unfold all such hidden discussions at once, so you can catch up.

Checking out pull requests as remote-tracking branches#

In your local fork, add to your.git/config, under the[remote"upstream"] heading, the line:

fetch=+refs/pull/*/head:refs/remotes/upstream/pr/*

You may then usegitcheckoutpr/PR_NUMBER to navigate to the code of thepull-request with the given number. (Read more in this gist.)

Display code coverage in pull requests#

To overlay the code coverage reports generated by the CodeCov continuousintegration, considerthis browser extension. The coverage of each linewill be displayed as a color background behind the line number.

Useful pytest aliases and flags#

The full test suite takes fairly long to run. For faster iterations,it is possible to select a subset of tests using pytest selectors.In particular, one can run asingle test based on its node ID:

pytest-vsklearn/linear_model/tests/test_logistic.py::test_sparsify

or use the-k pytest parameterto select tests based on their name. For instance,:

pytestsklearn/tests/test_common.py-v-kLogisticRegression

will run allcommon tests for theLogisticRegression estimator.

When a unit test fails, the following tricks can make debugging easier:

  1. The command line argumentpytest-l instructs pytest to print the localvariables when a failure occurs.

  2. The argumentpytest--pdb drops into the Python debugger on failure. Toinstead drop into the rich IPython debuggeripdb, you may set up ashell alias to:

    pytest--pdbcls=IPython.terminal.debugger:TerminalPdb--captureno

Otherpytest options that may become useful include:

  • -x which exits on the first failed test,

  • --lf to rerun the tests that failed on the previous run,

  • --ff to rerun all previous tests, running the ones that failed first,

  • -s so that pytest does not capture the output ofprint() statements,

  • --tb=short or--tb=line to control the length of the logs,

  • --runxfail also run tests marked as a known failure (XFAIL) and report errors.

Since our continuous integration tests will error ifFutureWarning isn’t properly caught,it is also recommended to runpytest along with the-Werror::FutureWarning flag.

Standard replies for reviewing#

It may be helpful to store some of these in GitHub’ssavedreplies for reviewing:

Issue: Usage questions

You are asking a usage question. The issue tracker is for bugs and new features. For usage questions, it is recommended to try [Stack Overflow](https://stackoverflow.com/questions/tagged/scikit-learn) or [the Mailing List](https://mail.python.org/mailman/listinfo/scikit-learn).Unfortunately, we need to close this issue as this issue tracker is a communication tool used for the development of scikit-learn. The additional activity created by usage questions crowds it too much and impedes this development. The conversation can continue here, however there is no guarantee that it will receive attention from core developers.

Issue: You’re welcome to update the docs

Please feel free to offer a pull request updating the documentation if you feel it could be improved.

Issue: Self-contained example for bug

Please provide [self-contained example code](https://scikit-learn.org/dev/developers/minimal_reproducer.html), including imports and data (if possible), so that other contributors can just run it and reproduce your issue. Ideally your example code should be minimal.

Issue: Software versions

To help diagnose your issue, please paste the output of:```pyimport sklearn; sklearn.show_versions()```Thanks.

Issue: Code blocks

Readability can be greatly improved if you [format](https://help.github.com/articles/creating-and-highlighting-code-blocks/) your code snippets and complete error messages appropriately. For example:    ```python    print(something)    ```generates:```pythonprint(something)```And:    ```pytb    Traceback (most recent call last):        File "<stdin>", line 1, in <module>    ImportError: No module named 'hello'    ```generates:```pytbTraceback (most recent call last):    File "<stdin>", line 1, in <module>ImportError: No module named 'hello'```You can edit your issue descriptions and comments at any time to improve readability. This helps maintainers a lot. Thanks!

Issue/Comment: Linking to code

Friendly advice: for clarity's sake, you can link to code like [this](https://help.github.com/articles/creating-a-permanent-link-to-a-code-snippet/).

Issue/Comment: Linking to comments

Please use links to comments, which make it a lot easier to see what you are referring to, rather than just linking to the issue. See [this](https://stackoverflow.com/questions/25163598/how-do-i-reference-a-specific-issue-comment-on-github) for more details.

PR-NEW: Better description and title

Thanks for the pull request! Please make the title of the PR more descriptive. The title will become the commit message when this is merged. You should state what issue (or PR) it fixes/resolves in the description using the syntax described [here](https://scikit-learn.org/dev/developers/contributing.html#contributing-pull-requests).

PR-NEW: Fix #

Please use "Fix #issueNumber" in your PR description (and you can do it more than once). This way the associated issue gets closed automatically when the PR is merged. For more details, look at [this](https://github.com/blog/1506-closing-issues-via-pull-requests).

PR-NEW or Issue: Maintenance cost

Every feature we include has a [maintenance cost](https://scikit-learn.org/dev/faq.html#why-are-you-so-selective-on-what-algorithms-you-include-in-scikit-learn). Our maintainers are mostly volunteers. For a new feature to be included, we need evidence that it is often useful and, ideally, [well-established](https://scikit-learn.org/dev/faq.html#what-are-the-inclusion-criteria-for-new-algorithms) in the literature or in practice. Also, we expect PR authors to take part in the maintenance for the code they submit, at least initially. That doesn't stop you implementing it for yourself and publishing it in a separate repository, or even [scikit-learn-contrib](https://scikit-learn-contrib.github.io).

PR-WIP: What’s needed before merge?

Please clarify (perhaps as a TODO list in the PR description) what work you believe still needs to be done before it can be reviewed for merge. When it is ready, please prefix the PR title with `[MRG]`.

PR-WIP: Regression test needed

Please add a [non-regression test](https://en.wikipedia.org/wiki/Non-regression_testing) that would fail at main but pass in this PR.

PR-MRG: Patience

Before merging, we generally require two core developers to agree that your pull request is desirable and ready. [Please be patient](https://scikit-learn.org/dev/faq.html#why-is-my-pull-request-not-getting-any-attention), as we mostly rely on volunteered time from busy core developers. (You are also welcome to help us out with [reviewing other PRs](https://scikit-learn.org/dev/developers/contributing.html#code-review-guidelines).)

PR-MRG: Add to what’s new

Please add an entry to the future changelog by adding an RST fragment into the module associated with your change located in `doc/whats_new/upcoming_changes`. Refer to the following [README](https://github.com/scikit-learn/scikit-learn/blob/main/doc/whats_new/upcoming_changes/README.md) for full instructions.

PR: Don’t change unrelated

Please do not change unrelated lines. It makes your contribution harder to review and may introduce merge conflicts to other pull requests.

Debugging CI issues#

CI issues may arise for a variety of reasons, so this is by no means acomprehensive guide, but rather a list of useful tips and tricks.

Using a lock-file to get an environment close to the CI#

conda-lock can be used to create a conda environment with the exact sameconda and pip packages as on the CI. For example, the following command willcreate a conda environment namedscikit-learn-doc that is similar to the CI:

conda-lockinstall-nscikit-learn-docbuild_tools/circle/doc_linux-64_conda.lock

Note

It only works if you have the same OS as the CI build (checkplatform: inthe lock-file). For example, the previous command will only work if you areon a Linux machine. Also this may not allow you to reproduce some of theissues that are more tied to the particularities of the CI environment, forexample CPU architecture reported by OpenBLAS insklearn.show_versions().

If you don’t have the same OS as the CI build you can still create a condaenvironment from the right environment yaml file, although it won’t be as closeas the CI environment as using the associated lock-file. For example for thedoc build:

condaenvcreate-nscikit-learn-doc-fbuild_tools/circle/doc_environment.yml-y

This may not give you exactly the same package versions as in the CI for avariety of reasons, for example:

  • some packages may have had new releases between the time the lock files werelast updated in themain branch and the time you run thecondacreatecommand. You can always try to look at the version in the lock-file andspecify the versions by hand for some specific packages that you think wouldhelp reproducing the issue.

  • different packages may be installed by default depending on the OS. Forexample, the default BLAS library when installing numpy is OpenBLAS on Linuxand MKL on Windows.

Also the problem may be OS specific so the only way to be able to reproducewould be to have the same OS as the CI build.

Debugging memory errors in Cython with valgrind#

While python/numpy’s built-in memory management is relatively robust, it canlead to performance penalties for some routines. For this reason, much ofthe high-performance code in scikit-learn is written in cython. Thisperformance gain comes with a tradeoff, however: it is very easy for memorybugs to crop up in cython code, especially in situations where that coderelies heavily on pointer arithmetic.

Memory errors can manifest themselves a number of ways. The easiest ones todebug are often segmentation faults and related glibc errors. Uninitializedvariables can lead to unexpected behavior that is difficult to track down.A very useful tool when debugging these sorts of errors isvalgrind.

Valgrind is a command-line tool that can trace memory errors in a variety ofcode. Follow these steps:

  1. Installvalgrind on your system.

  2. Download the python valgrind suppression file:valgrind-python.supp.

  3. Follow the directions in theREADME.valgrind file to customize yourpython suppressions. If you don’t, you will have spurious output comingrelated to the python interpreter instead of your own code.

  4. Run valgrind as follows:

    valgrind-v--suppressions=valgrind-python.supppythonmy_test_script.py

The result will be a list of all the memory-related errors, which referencelines in the C-code generated by cython from your .pyx file. If you examinethe referenced lines in the .c file, you will see comments which indicate thecorresponding location in your .pyx source file. Hopefully the output willgive you clues as to the source of your memory error.

For more information on valgrind and the array of options it has, see thetutorials and documentation on thevalgrind web site.

Building and testing for the ARM64 platform on a x86_64 machine#

ARM-based machines are a popular target for mobile, edge or other low-energydeployments (including in the cloud, for instance on Scaleway or AWS Graviton).

Here are instructions to setup a local dev environment to reproduceARM-specific bugs or test failures on a x86_64 host laptop or workstation. Thisis based on QEMU user mode emulation using docker for convenience (seemultiarch/qemu-user-static).

Note

The following instructions are illustrated for ARM64 but they also apply toppc64le, after changing the Docker image and Miniforge paths appropriately.

Prepare a folder on the host filesystem and download the necessary tools andsource code:

mkdirarm64pushdarm64wgethttps://github.com/conda-forge/miniforge/releases/latest/download/Miniforge3-Linux-aarch64.shgitclonehttps://github.com/scikit-learn/scikit-learn.git

Use docker to install QEMU user mode and run an ARM64v8 container with accessto your shared folder under the/io mount point:

dockerrun--rm--privilegedmultiarch/qemu-user-static--reset-pyesdockerrun-v`pwd`:/io--rm-itarm64v8/ubuntu/bin/bash

In the container, install miniforge3 for the ARM64 (a.k.a. aarch64)architecture:

bashMiniforge3-Linux-aarch64.sh# Choose to install miniforge3 under: `/io/miniforge3`

Whenever you restart a new container, you will need to reinit the conda envpreviously installed under/io/miniforge3:

/io/miniforge3/bin/condainitsource/root/.bashrc

as the/root home folder is part of the ephemeral docker container. Everyfile or directory stored under/io is persistent on the other hand.

You can then build scikit-learn as usual (you will need to install compilertools and dependencies using apt or conda as usual). Building scikit-learntakes a lot of time because of the emulation layer, however it needs to bedone only once if you put the scikit-learn folder under the/io mountpoint.

Then use pytest to run only the tests of the module you are interested indebugging.

The Meson Build Backend#

Since scikit-learn 1.5.0 we use meson-python as the build tool. Meson isa new tool for scikit-learn and the PyData ecosystem. It is used by severalother packages that have written good guides about what it is and how it works.