Contributing to pandas¶

Table of contents:

Where to start?
Bug reports and enhancement requests
Working with the code
Contributing to the documentation
Contributing to the code base
Contributing your changes to pandas
Tips for a successful Pull Request

Where to start?¶

All contributions, bug reports, bug fixes, documentation improvements, enhancements, and ideas are welcome.

If you are brand new to pandas or open-source development, we recommend going through the GitHub “issues” tab to find issues that interest you. There are a number of issues listed under Docs and good first issue where you could start out. Once you’ve found an interesting issue, you can return here to get your development environment setup.

When you start working on an issue, it’s a good idea to assign the issue to yourself, so nobody else duplicates the work on it. GitHub restricts assigning issues to maintainers of the project only. In most projects, and until recently in pandas, contributors added a comment letting others know they are working on an issue. While this is ok, you need to check each issue individually, and it’s not possible to find the unassigned ones.

For this reason, we implemented a workaround consisting of adding a comment with the exact text take. When you do it, a GitHub action will automatically assign you the issue (this will take seconds, and may require refreshint the page to see it). By doing this, it’s possible to filter the list of issues and find only the unassigned ones.

So, a good way to find an issue to start contributing to pandas is to check the list of unassigned good first issues and assign yourself one you like by writing a comment with the exact text take.

If for whatever reason you are not able to continue working with the issue, please try to unassign it, so other people know it’s available again. You can check the list of assigned issues, since people may not be working in them anymore. If you want to work on one that is assigned, feel free to kindly ask the current assignee if you can take it (please allow at least a week of inactivity before considering work in the issue discontinued).

Feel free to ask questions on the mailing list or on Gitter.

Bug reports and enhancement requests ¶

Bug reports are an important part of making pandas more stable. Having a complete bug report will allow others to reproduce the bug and provide insight into fixing. See this stackoverflow article and this blogpost for tips on writing a good bug report.

Trying the bug-producing code out on the master branch is often a worthwhile exercise to confirm the bug still exists. It is also worth searching existing bug reports and pull requests to see if the issue has already been reported and/or fixed.

Bug reports must:

Include a short, self-contained Python snippet reproducing the problem. You can format the code nicely by using GitHub Flavored Markdown:
```
```python
>>> from pandas import DataFrame
>>> df = DataFrame(...)
...
```
```
Include the full version string of pandas and its dependencies. You can use the built-in function:
```
>>> import pandas as pd
>>> pd.show_versions()
```
Explain why the current behavior is wrong/not desired and what you expect instead.

The issue will then show up to the pandas community and be open to comments/ideas from others.

Working with the code ¶

Now that you have an issue you want to fix, enhancement to add, or documentation to improve, you need to learn how to work with GitHub and the pandas code base.

Version control, Git, and GitHub ¶

To the new user, working with Git is one of the more daunting aspects of contributing to pandas. It can very quickly become overwhelming, but sticking to the guidelines below will help keep the process straightforward and mostly trouble free. As always, if you are having difficulties please feel free to ask for help.

The code is hosted on GitHub. To contribute you will need to sign up for a free GitHub account. We use Git for version control to allow many people to work together on the project.

Some great resources for learning Git:

the GitHub help pages.
the NumPy’s documentation.
Matthew Brett’s Pydagogue.

Getting started with Git ¶

GitHub has instructions for installing git, setting up your SSH key, and configuring git. All these steps need to be completed before you can work seamlessly between your local repository and GitHub.

Forking ¶

You will need your own fork to work on the code. Go to the pandas project page and hit the Fork button. You will want to clone your fork to your machine:

git clone https://github.com/your-user-name/pandas.git pandas-yourname
cd pandas-yourname
git remote add upstream https://github.com/pandas-dev/pandas.git

This creates the directory pandas-yourname and connects your repository to the upstream (main project) pandas repository.

Creating a development environment ¶

To test out code changes, you’ll need to build pandas from source, which requires a C compiler and Python environment. If you’re making documentation changes, you can skip to Contributing to the documentation but you won’t be able to build the documentation locally before pushing your changes.

Using a Docker Container ¶

Instead of manually setting up a development environment, you can use Docker to automatically create the environment with just several commands. Pandas provides a DockerFile in the root directory to build a Docker image with a full pandas development environment.

Even easier, you can use the DockerFile to launch a remote session with Visual Studio Code, a popular free IDE, using the .devcontainer.json file. See https://code.visualstudio.com/docs/remote/containers for details.

Installing a C compiler ¶

Pandas uses C extensions (mostly written using Cython) to speed up certain operations. To install pandas from source, you need to compile these C extensions, which means you need a C compiler. This process depends on which platform you’re using.

Windows

You will need Build Tools for Visual Studio 2017.

Warning

You DO NOT need to install Visual Studio 2019. You only need “Build Tools for Visual Studio 2019” found by scrolling down to “All downloads” -> “Tools for Visual Studio 2019”.

Mac OS

Information about compiler installation can be found here: https://devguide.python.org/setup/#macos

Unix

Some Linux distributions will come with a pre-installed C compiler. To find out which compilers (and versions) are installed on your system:

# for Debian/Ubuntu:
dpkg --list | grep compiler
# for Red Hat/RHEL/CentOS/Fedora:
yum list installed | grep -i --color compiler

GCC (GNU Compiler Collection), is a widely used compiler, which supports C and a number of other languages. If GCC is listed as an installed compiler nothing more is required. If no C compiler is installed (or you wish to install a newer version) you can install a compiler (GCC in the example code below) with:

# for recent Debian/Ubuntu:
sudo apt install build-essential
# for Red Had/RHEL/CentOS/Fedora
yum groupinstall "Development Tools"

For other Linux distributions, consult your favourite search engine for compiler installation instructions.

Let us know if you have any difficulties by opening an issue or reaching out on Gitter.

Creating a Python environment ¶

Now that you have a C compiler, create an isolated pandas development environment:

Install either Anaconda or miniconda
Make sure your conda is up to date (conda update conda)
Make sure that you have cloned the repository
cd to the pandas source directory

We’ll now kick off a three-step process:

Install the build dependencies
Build and install pandas
Install the optional dependencies

# Create and activate the build environment
conda env create -f environment.yml
conda activate pandas-dev

# or with older versions of Anaconda:
source activate pandas-dev

# Build and install pandas
python setup.py build_ext --inplace -j 4
python -m pip install -e . --no-build-isolation --no-use-pep517

At this point you should be able to import pandas from your locally built version:

$ python  # start an interpreter
>>> import pandas
>>> print(pandas.__version__)
0.22.0.dev0+29.g4ad6d4d74

This will create the new environment, and not touch any of your existing environments, nor any existing Python installation.

To view your environments:

conda info -e

To return to your root environment:

conda deactivate

See the full conda docs here.

Creating a Python environment (pip)¶

If you aren’t using conda for your development environment, follow these instructions. You’ll need to have at least Python 3.6.1 installed on your system.

Unix/Mac OS

# Create a virtual environment
# Use an ENV_DIR of your choice. We'll use ~/virtualenvs/pandas-dev
# Any parent directories should already exist
python3 -m venv ~/virtualenvs/pandas-dev

# Activate the virtualenv
. ~/virtualenvs/pandas-dev/bin/activate

# Install the build dependencies
python -m pip install -r requirements-dev.txt

# Build and install pandas
python setup.py build_ext --inplace -j 0
python -m pip install -e . --no-build-isolation --no-use-pep517

Windows

Below is a brief overview on how to set-up a virtual environment with Powershell under Windows. For details please refer to the official virtualenv user guide

Use an ENV_DIR of your choice. We’ll use ~virtualenvspandas-dev where ‘~’ is the folder pointed to by either $env:USERPROFILE (Powershell) or %USERPROFILE% (cmd.exe) environment variable. Any parent directories should already exist.

# Create a virtual environment
python -m venv $env:USERPROFILE\virtualenvs\pandas-dev

# Activate the virtualenv. Use activate.bat for cmd.exe
~\virtualenvs\pandas-dev\Scripts\Activate.ps1

# Install the build dependencies
python -m pip install -r requirements-dev.txt

# Build and install pandas
python setup.py build_ext --inplace -j 0
python -m pip install -e . --no-build-isolation --no-use-pep517

Creating a branch ¶

You want your master branch to reflect only production-ready code, so create a feature branch for making your changes. For example:

git branch shiny-new-feature
git checkout shiny-new-feature

The above can be simplified to:

git checkout -b shiny-new-feature

This changes your working directory to the shiny-new-feature branch. Keep any changes in this branch specific to one bug or feature so it is clear what the branch brings to pandas. You can have many shiny-new-features and switch in between them using the git checkout command.

When creating this branch, make sure your master branch is up to date with the latest upstream master version. To update your local master branch, you can do:

git checkout master
git pull upstream master --ff-only

When you want to update the feature branch with changes in master after you created the branch, check the section on updating a PR.

Contributing to the documentation ¶

Contributing to the documentation benefits everyone who uses pandas. We encourage you to help us improve the documentation, and you don’t have to be an expert on pandas to do so! In fact, there are sections of the docs that are worse off after being written by experts. If something in the docs doesn’t make sense to you, updating the relevant section after you figure it out is a great way to ensure it will help the next person.

Documentation:

About the pandas documentation
Updating a pandas docstring
How to build the pandas documentation

About the pandas documentation ¶

The documentation is written in reStructuredText, which is almost like writing in plain English, and built using Sphinx. The Sphinx Documentation has an excellent introduction to reST. Review the Sphinx docs to perform more complex changes to the documentation as well.

Some other important things to know about the docs:

The pandas documentation consists of two parts: the docstrings in the code itself and the docs in this folder doc/.

The docstrings provide a clear explanation of the usage of the individual functions, while the documentation in this folder consists of tutorial-like overviews per topic together with some other information (what’s new, installation, etc).
The docstrings follow a pandas convention, based on the Numpy Docstring Standard. Follow the pandas docstring guide for detailed instructions on how to write a correct docstring.
- pandas docstring guide
The tutorials make heavy use of the ipython directive sphinx extension. This directive lets you put code in the documentation which will be run during the doc build. For example:
```
.. ipython:: python

    x = 2
    x**3
```
will be rendered as:
```
In [1]: x = 2

In [2]: x**3
Out[2]: 8
```
Almost all code examples in the docs are run (and the output saved) during the doc build. This approach means that code examples will always be up to date, but it does make the doc building a bit more complex.
Our API documentation in doc/source/api.rst houses the auto-generated documentation from the docstrings. For classes, there are a few subtleties around controlling which methods and attributes have pages auto-generated.

We have two autosummary templates for classes.
1. _templates/autosummary/class.rst. Use this when you want to automatically generate a page for every public method and attribute on the class. The Attributes and Methods sections will be automatically added to the class’ rendered documentation by numpydoc. See DataFrame for an example.
2. _templates/autosummary/class_without_autosummary. Use this when you want to pick a subset of methods / attributes to auto-generate pages for. When using this template, you should include an Attributes and Methods section in the class docstring. See CategoricalIndex for an example.
Every method should be included in a toctree in api.rst, else Sphinx will emit a warning.

Note

The .rst files are used to automatically generate Markdown and HTML versions of the docs. For this reason, please do not edit CONTRIBUTING.md directly, but instead make any changes to doc/source/development/contributing.rst. Then, to generate CONTRIBUTING.md, use pandoc with the following command:

pandoc doc/source/development/contributing.rst -t markdown_github > CONTRIBUTING.md

The utility script scripts/validate_docstrings.py can be used to get a csv summary of the API documentation. And also validate common errors in the docstring of a specific class, function or method. The summary also compares the list of methods documented in doc/source/api.rst (which is used to generate the API Reference page) and the actual public methods. This will identify methods documented in doc/source/api.rst that are not actually class methods, and existing methods that are not documented in doc/source/api.rst.

Updating a pandas docstring ¶

When improving a single function or method’s docstring, it is not necessarily needed to build the full documentation (see next section). However, there is a script that checks a docstring (for example for the DataFrame.mean method):

python scripts/validate_docstrings.py pandas.DataFrame.mean

This script will indicate some formatting errors if present, and will also run and test the examples included in the docstring. Check the pandas docstring guide for a detailed guide on how to format the docstring.

The examples in the docstring (‘doctests’) must be valid Python code, that in a deterministic way returns the presented output, and that can be copied and run by users. This can be checked with the script above, and is also tested on Travis. A failing doctest will be a blocker for merging a PR. Check the examples section in the docstring guide for some tips and tricks to get the doctests passing.

When doing a PR with a docstring update, it is good to post the output of the validation script in a comment on github.

How to build the pandas documentation ¶

Requirements ¶

First, you need to have a development environment to be able to build pandas (see the docs on creating a development environment above).

Building the documentation ¶

So how do you build the docs? Navigate to your local doc/ directory in the console and run:

python make.py html

Then you can find the HTML output in the folder doc/build/html/.

The first time you build the docs, it will take quite a while because it has to run all the code examples and build all the generated docstring pages. In subsequent evocations, sphinx will try to only build the pages that have been modified.

If you want to do a full clean build, do:

python make.py clean
python make.py html

You can tell make.py to compile only a single section of the docs, greatly reducing the turn-around time for checking your changes.

# omit autosummary and API section
python make.py clean
python make.py --no-api

# compile the docs with only a single section, relative to the "source" folder.
# For example, compiling only this guide (doc/source/development/contributing.rst)
python make.py clean
python make.py --single development/contributing.rst

# compile the reference docs for a single function
python make.py clean
python make.py --single pandas.DataFrame.join

For comparison, a full documentation build may take 15 minutes, but a single section may take 15 seconds. Subsequent builds, which only process portions you have changed, will be faster.

You can also specify to use multiple cores to speed up the documentation build:

python make.py html --num-jobs 4

Open the following file in a web browser to see the full documentation you just built:

doc/build/html/index.html

And you’ll have the satisfaction of seeing your new and improved documentation!

Building master branch documentation ¶

When pull requests are merged into the pandas master branch, the main parts of the documentation are also built by Travis-CI. These docs are then hosted here, see also the Continuous Integration section.

Contributing to the code base ¶

Code Base:

Code standards ¶

Writing good code is not just about what you write. It is also about how you write it. During Continuous Integration testing, several tools will be run to check your code for stylistic errors. Generating any warnings will cause the test to fail. Thus, good style is a requirement for submitting code to pandas.

There is a tool in pandas to help contributors verify their changes before contributing them to the project:

./ci/code_checks.sh

The script verifies the linting of code files, it looks for common mistake patterns (like missing spaces around sphinx directives that make the documentation not being rendered properly) and it also validates the doctests. It is possible to run the checks independently by using the parameters lint, patterns and doctests (e.g. ./ci/code_checks.sh lint).

In addition, because a lot of people use our library, it is important that we do not make sudden changes to the code that could have the potential to break a lot of user code as a result, that is, we need it to be as backwards compatible as possible to avoid mass breakages.

Additional standards are outlined on the pandas code style guide

Optional dependencies ¶

Optional dependencies (e.g. matplotlib) should be imported with the private helper pandas.compat._optional.import_optional_dependency. This ensures a consistent error message when the dependency is not met.

All methods using an optional dependency should include a test asserting that an ImportError is raised when the optional dependency is not found. This test should be skipped if the library is present.

All optional dependencies should be documented in Optional dependencies and the minimum required version should be set in the pandas.compat._optional.VERSIONS dict.

C (cpplint)¶

pandas uses the Google standard. Google provides an open source style checker called cpplint, but we use a fork of it that can be found here. Here are some of the more common cpplint issues:

we restrict line-length to 80 characters to promote readability
every header file must include a header guard to avoid name collisions if re-included

Continuous Integration will run the cpplint tool and report any stylistic errors in your code. Therefore, it is helpful before submitting code to run the check yourself:

cpplint --extensions=c,h --headers=h --filter=-readability/casting,-runtime/int,-build/include_subdir modified-c-file

You can also run this command on an entire directory if necessary:

cpplint --extensions=c,h --headers=h --filter=-readability/casting,-runtime/int,-build/include_subdir --recursive modified-c-directory

To make your commits compliant with this standard, you can install the ClangFormat tool, which can be downloaded here. To configure, in your home directory, run the following command:

clang-format style=google -dump-config  > .clang-format

Then modify the file to ensure that any indentation width parameters are at least four. Once configured, you can run the tool as follows:

clang-format modified-c-file

This will output what your file will look like if the changes are made, and to apply them, run the following command:

clang-format -i modified-c-file

To run the tool on an entire directory, you can run the following analogous commands:

clang-format modified-c-directory/*.c modified-c-directory/*.h
clang-format -i modified-c-directory/*.c modified-c-directory/*.h

Do note that this tool is best-effort, meaning that it will try to correct as many errors as possible, but it may not correct all of them. Thus, it is recommended that you run cpplint to double check and make any other style fixes manually.

Python (PEP8 / black)¶

pandas follows the PEP8 standard and uses Black and Flake8 to ensure a consistent code format throughout the project.

Continuous Integration will run those tools and report any stylistic errors in your code. Therefore, it is helpful before submitting code to run the check yourself:

black pandas
git diff upstream/master -u -- "*.py" | flake8 --diff

to auto-format your code. Additionally, many editors have plugins that will apply black as you edit files.

You should use a black version >= 19.10b0 as previous versions are not compatible with the pandas codebase.

If you wish to run these checks automatically, we encourage you to use pre-commits instead.

One caveat about git diff upstream/master -u -- "*.py" | flake8 --diff: this command will catch any stylistic errors in your changes specifically, but be beware it may not catch all of them. For example, if you delete the only usage of an imported function, it is stylistically incorrect to import an unused function. However, style-checking the diff will not catch this because the actual import is not part of the diff. Thus, for completeness, you should run this command, though it may take longer:

git diff upstream/master --name-only -- "*.py" | xargs -r flake8

Note that on OSX, the -r flag is not available, so you have to omit it and run this slightly modified command:

git diff upstream/master --name-only -- "*.py" | xargs flake8

Windows does not support the xargs command (unless installed for example via the MinGW toolchain), but one can imitate the behaviour as follows:

for /f %i in ('git diff upstream/master --name-only -- "*.py"') do flake8 %i

This will get all the files being changed by the PR (and ending with .py), and run flake8 on them, one after the other.

Note that these commands can be run analogously with black.

Import formatting ¶

pandas uses isort to standardise import formatting across the codebase.

A guide to import layout as per pep8 can be found here.

A summary of our current import sections ( in order ):

Future
Python Standard Library
Third Party
pandas._libs, pandas.compat, pandas.util._*, pandas.errors (largely not dependent on pandas.core)
pandas.core.dtypes (largely not dependent on the rest of pandas.core)
Rest of pandas.core.*
Non-core pandas.io, pandas.plotting, pandas.tseries
Local application/library specific imports

Imports are alphabetically sorted within these sections.

As part of Continuous Integration checks we run:

isort --recursive --check-only pandas

to check that imports are correctly formatted as per the setup.cfg.

If you see output like the below in Continuous Integration checks:

Check import format using isort
ERROR: /home/travis/build/pandas-dev/pandas/pandas/io/pytables.py Imports are incorrectly sorted
Check import format using isort DONE
The command "ci/code_checks.sh" exited with 1

You should run:

isort pandas/io/pytables.py

to automatically format imports correctly. This will modify your local copy of the files.

The –recursive flag can be passed to sort all files in a directory.

Alternatively, you can run a command similar to what was suggested for black and flake8 right above:

git diff upstream/master --name-only -- "*.py" | xargs -r isort

Where similar caveats apply if you are on OSX or Windows.

You can then verify the changes look ok, then git commit and push.

Pre-Commit ¶

You can run many of these styling checks manually as we have described above. However, we encourage you to use pre-commit hooks instead to automatically run black, flake8, isort when you make a git commit. This can be done by installing pre-commit:

pip install pre-commit

and then running:

pre-commit install

from the root of the pandas repository. Now all of the styling checks will be run each time you commit changes without your needing to run each one manually. In addition, using this pre-commit hook will also allow you to more easily remain up-to-date with our code checks as they change.

Note that if needed, you can skip these checks with git commit --no-verify.

Backwards compatibility ¶

Please try to maintain backward compatibility. pandas has lots of users with lots of existing code, so don’t break it if at all possible. If you think breakage is required, clearly state why as part of the pull request. Also, be careful when changing method signatures and add deprecation warnings where needed. Also, add the deprecated sphinx directive to the deprecated functions or methods.

If a function with the same arguments as the one being deprecated exist, you can use the pandas.util._decorators.deprecate:

from pandas.util._decorators import deprecate

deprecate('old_func', 'new_func', '0.21.0')

Otherwise, you need to do it manually:

import warnings


def old_func():
    """Summary of the function.

    .. deprecated:: 0.21.0
       Use new_func instead.
    """
    warnings.warn('Use new_func instead.', FutureWarning, stacklevel=2)
    new_func()


def new_func():
    pass

You’ll also need to

Write a new test that asserts a warning is issued when calling with the deprecated argument
Update all of pandas existing tests and code to use the new argument

See Testing warnings for more.

Type Hints ¶

pandas strongly encourages the use of PEP 484 style type hints. New development should contain type hints and pull requests to annotate existing code are accepted as well!

Style Guidelines ¶

Types imports should follow the from typing import ... convention. So rather than

import typing

primes: typing.List[int] = []

You should write

from typing import List, Optional, Union

primes: List[int] = []

Optional should be used where applicable, so instead of

maybe_primes: List[Union[int, None]] = []

You should write

maybe_primes: List[Optional[int]] = []

In some cases in the code base classes may define class variables that shadow builtins. This causes an issue as described in Mypy 1775. The defensive solution here is to create an unambiguous alias of the builtin and use that without your annotation. For example, if you come across a definition like

class SomeClass1:
    str = None

The appropriate way to annotate this would be as follows

str_type = str

class SomeClass2:
    str: str_type = None

In some cases you may be tempted to use cast from the typing module when you know better than the analyzer. This occurs particularly when using custom inference functions. For example

from typing import cast

from pandas.core.dtypes.common import is_number

def cannot_infer_bad(obj: Union[str, int, float]):

    if is_number(obj):
        ...
    else:  # Reasonably only str objects would reach this but...
        obj = cast(str, obj)  # Mypy complains without this!
        return obj.upper()

The limitation here is that while a human can reasonably understand that is_number would catch the int and float types mypy cannot make that same inference just yet (see mypy #5206. While the above works, the use of cast is strongly discouraged. Where applicable a refactor of the code to appease static analysis is preferable

def cannot_infer_good(obj: Union[str, int, float]):

    if isinstance(obj, str):
        return obj.upper()
    else:
        ...

With custom types and inference this is not always possible so exceptions are made, but every effort should be exhausted to avoid cast before going down such paths.

Pandas-specific Types ¶

Commonly used types specific to pandas will appear in pandas._typing and you should use these where applicable. This module is private for now but ultimately this should be exposed to third party libraries who want to implement type checking against pandas.

For example, quite a few functions in pandas accept a dtype argument. This can be expressed as a string like "object", a numpy.dtype like np.int64 or even a pandas ExtensionDtype like pd.CategoricalDtype. Rather than burden the user with having to constantly annotate all of those options, this can simply be imported and reused from the pandas._typing module

from pandas._typing import Dtype

def as_type(dtype: Dtype) -> ...:
    ...

This module will ultimately house types for repeatedly used concepts like “path-like”, “array-like”, “numeric”, etc… and can also hold aliases for commonly appearing parameters like axis. Development of this module is active so be sure to refer to the source for the most up to date list of available types.

Validating Type Hints ¶

pandas uses mypy to statically analyze the code base and type hints. After making any change you can ensure your type hints are correct by running

mypy pandas

Testing with continuous integration ¶

The pandas test suite will run automatically on Travis-CI and Azure Pipelines continuous integration services, once your pull request is submitted. However, if you wish to run the test suite on a branch prior to submitting the pull request, then the continuous integration services need to be hooked to your GitHub repository. Instructions are here for Travis-CI and Azure Pipelines.

A pull-request will be considered for merging when you have an all ‘green’ build. If any tests are failing, then you will get a red ‘X’, where you can click through to see the individual failed tests. This is an example of a green build.

Note

Each time you push to your fork, a new run of the tests will be triggered on the CI. You can enable the auto-cancel feature, which removes any non-currently-running tests for that same pull-request, for Travis-CI here.

Test-driven development/code writing ¶

pandas is serious about testing and strongly encourages contributors to embrace test-driven development (TDD). This development process “relies on the repetition of a very short development cycle: first the developer writes an (initially failing) automated test case that defines a desired improvement or new function, then produces the minimum amount of code to pass that test.” So, before actually writing any code, you should write your tests. Often the test can be taken from the original GitHub issue. However, it is always worth considering additional use cases and writing corresponding tests.

Adding tests is one of the most common requests after code is pushed to pandas. Therefore, it is worth getting in the habit of writing tests ahead of time so this is never an issue.

Like many packages, pandas uses pytest and the convenient extensions in numpy.testing.

Note

The earliest supported pytest version is 5.0.1.

Writing tests ¶

All tests should go into the tests subdirectory of the specific package. This folder contains many current examples of tests, and we suggest looking to these for inspiration. If your test requires working with files or network connectivity, there is more information on the testing page of the wiki.

The pandas._testing module has many special assert functions that make it easier to make statements about whether Series or DataFrame objects are equivalent. The easiest way to verify that your code is correct is to explicitly construct the result you expect, then compare the actual result to the expected correct result:

def test_pivot(self):
    data = {
        'index' : ['A', 'B', 'C', 'C', 'B', 'A'],
        'columns' : ['One', 'One', 'One', 'Two', 'Two', 'Two'],
        'values' : [1., 2., 3., 3., 2., 1.]
    }

    frame = DataFrame(data)
    pivoted = frame.pivot(index='index', columns='columns', values='values')

    expected = DataFrame({
        'One' : {'A' : 1., 'B' : 2., 'C' : 3.},
        'Two' : {'A' : 1., 'B' : 2., 'C' : 3.}
    })

    assert_frame_equal(pivoted, expected)

Please remember to add the Github Issue Number as a comment to a new test. E.g. “# brief comment, see GH#28907”

Transitioning to `pytest`¶

pandas existing test structure is mostly class-based, meaning that you will typically find tests wrapped in a class.

class TestReallyCoolFeature:
    pass

Going forward, we are moving to a more functional style using the pytest framework, which offers a richer testing framework that will facilitate testing and developing. Thus, instead of writing test classes, we will write test functions like this:

def test_really_cool_feature():
    pass

Using `pytest`¶

Here is an example of a self-contained set of tests that illustrate multiple features that we like to use.

functional style: tests are like test_* and only take arguments that are either fixtures or parameters
pytest.mark can be used to set metadata on test functions, e.g. skip or xfail.
using parametrize: allow testing of multiple cases
to set a mark on a parameter, pytest.param(..., marks=...) syntax should be used
fixture, code for object construction, on a per-test basis
using bare assert for scalars and truth-testing
tm.assert_series_equal (and its counter part tm.assert_frame_equal), for pandas object comparisons.
the typical pattern of constructing an expected and comparing versus the result

We would name this file test_cool_feature.py and put in an appropriate place in the pandas/tests/ structure.

import pytest
import numpy as np
import pandas as pd


@pytest.mark.parametrize('dtype', ['int8', 'int16', 'int32', 'int64'])
def test_dtypes(dtype):
    assert str(np.dtype(dtype)) == dtype


@pytest.mark.parametrize(
    'dtype', ['float32', pytest.param('int16', marks=pytest.mark.skip),
              pytest.param('int32', marks=pytest.mark.xfail(
                  reason='to show how it works'))])
def test_mark(dtype):
    assert str(np.dtype(dtype)) == 'float32'


@pytest.fixture
def series():
    return pd.Series([1, 2, 3])


@pytest.fixture(params=['int8', 'int16', 'int32', 'int64'])
def dtype(request):
    return request.param


def test_series(series, dtype):
    result = series.astype(dtype)
    assert result.dtype == dtype

    expected = pd.Series([1, 2, 3], dtype=dtype)
    tm.assert_series_equal(result, expected)

A test run of this yields

((pandas) bash-3.2$ pytest  test_cool_feature.py  -v
=========================== test session starts ===========================
platform darwin -- Python 3.6.2, pytest-3.6.0, py-1.4.31, pluggy-0.4.0
collected 11 items

tester.py::test_dtypes[int8] PASSED
tester.py::test_dtypes[int16] PASSED
tester.py::test_dtypes[int32] PASSED
tester.py::test_dtypes[int64] PASSED
tester.py::test_mark[float32] PASSED
tester.py::test_mark[int16] SKIPPED
tester.py::test_mark[int32] xfail
tester.py::test_series[int8] PASSED
tester.py::test_series[int16] PASSED
tester.py::test_series[int32] PASSED
tester.py::test_series[int64] PASSED

Tests that we have parametrized are now accessible via the test name, for example we could run these with -k int8 to sub-select only those tests which match int8.

((pandas) bash-3.2$ pytest  test_cool_feature.py  -v -k int8
=========================== test session starts ===========================
platform darwin -- Python 3.6.2, pytest-3.6.0, py-1.4.31, pluggy-0.4.0
collected 11 items

test_cool_feature.py::test_dtypes[int8] PASSED
test_cool_feature.py::test_series[int8] PASSED

Using `hypothesis`¶

Hypothesis is a library for property-based testing. Instead of explicitly parametrizing a test, you can describe all valid inputs and let Hypothesis try to find a failing input. Even better, no matter how many random examples it tries, Hypothesis always reports a single minimal counterexample to your assertions - often an example that you would never have thought to test.

See Getting Started with Hypothesis for more of an introduction, then refer to the Hypothesis documentation for details.

import json
from hypothesis import given, strategies as st

any_json_value = st.deferred(lambda: st.one_of(
    st.none(), st.booleans(), st.floats(allow_nan=False), st.text(),
    st.lists(any_json_value), st.dictionaries(st.text(), any_json_value)
))


@given(value=any_json_value)
def test_json_roundtrip(value):
    result = json.loads(json.dumps(value))
    assert value == result

This test shows off several useful features of Hypothesis, as well as demonstrating a good use-case: checking properties that should hold over a large or complicated domain of inputs.

To keep the Pandas test suite running quickly, parametrized tests are preferred if the inputs or logic are simple, with Hypothesis tests reserved for cases with complex logic or where there are too many combinations of options or subtle interactions to test (or think of!) all of them.

Testing warnings ¶

By default, one of pandas CI workers will fail if any unhandled warnings are emitted.

If your change involves checking that a warning is actually emitted, use tm.assert_produces_warning(ExpectedWarning).

import pandas._testing as tm


df = pd.DataFrame()
with tm.assert_produces_warning(FutureWarning):
    df.some_operation()

We prefer this to the pytest.warns context manager because ours checks that the warning’s stacklevel is set correctly. The stacklevel is what ensure the user’s file name and line number is printed in the warning, rather than something internal to pandas. It represents the number of function calls from user code (e.g. df.some_operation()) to the function that actually emits the warning. Our linter will fail the build if you use pytest.warns in a test.

If you have a test that would emit a warning, but you aren’t actually testing the warning itself (say because it’s going to be removed in the future, or because we’re matching a 3rd-party library’s behavior), then use pytest.mark.filterwarnings to ignore the error.

@pytest.mark.filterwarnings("ignore:msg:category")
def test_thing(self):
    ...

If the test generates a warning of class category whose message starts with msg, the warning will be ignored and the test will pass.

If you need finer-grained control, you can use Python’s usual warnings module to control whether a warning is ignored / raised at different places within a single test.

with warnings.catch_warnings():
    warnings.simplefilter("ignore", FutureWarning)
    # Or use warnings.filterwarnings(...)

Alternatively, consider breaking up the unit test.

Running the test suite ¶

The tests can then be run directly inside your Git clone (without having to install pandas) by typing:

pytest pandas

The tests suite is exhaustive and takes around 20 minutes to run. Often it is worth running only a subset of tests first around your changes before running the entire suite.

The easiest way to do this is with:

pytest pandas/path/to/test.py -k regex_matching_test_name

Or with one of the following constructs:

pytest pandas/tests/[test-module].py
pytest pandas/tests/[test-module].py::[TestClass]
pytest pandas/tests/[test-module].py::[TestClass]::[test_method]

Using pytest-xdist, one can speed up local testing on multicore machines. To use this feature, you will need to install pytest-xdist via:

pip install pytest-xdist

Two scripts are provided to assist with this. These scripts distribute testing across 4 threads.

On Unix variants, one can type:

test_fast.sh

On Windows, one can type:

test_fast.bat

This can significantly reduce the time it takes to locally run tests before submitting a pull request.

For more, see the pytest documentation.

Furthermore one can run

pd.test()

with an imported pandas to run tests similarly.

Running the performance test suite ¶

Performance matters and it is worth considering whether your code has introduced performance regressions. pandas is in the process of migrating to asv benchmarks to enable easy monitoring of the performance of critical pandas operations. These benchmarks are all found in the pandas/asv_bench directory. asv supports both python2 and python3.

To use all features of asv, you will need either conda or virtualenv. For more details please check the asv installation webpage.

To install asv:

pip install git+https://github.com/spacetelescope/asv

If you need to run a benchmark, change your directory to asv_bench/ and run:

asv continuous -f 1.1 upstream/master HEAD

You can replace HEAD with the name of the branch you are working on, and report benchmarks that changed by more than 10%. The command uses conda by default for creating the benchmark environments. If you want to use virtualenv instead, write:

asv continuous -f 1.1 -E virtualenv upstream/master HEAD

The -E virtualenv option should be added to all asv commands that run benchmarks. The default value is defined in asv.conf.json.

Running the full test suite can take up to one hour and use up to 3GB of RAM. Usually it is sufficient to paste only a subset of the results into the pull request to show that the committed changes do not cause unexpected performance regressions. You can run specific benchmarks using the -b flag, which takes a regular expression. For example, this will only run tests from a pandas/asv_bench/benchmarks/groupby.py file:

asv continuous -f 1.1 upstream/master HEAD -b ^groupby

If you want to only run a specific group of tests from a file, you can do it using . as a separator. For example:

asv continuous -f 1.1 upstream/master HEAD -b groupby.GroupByMethods

will only run the GroupByMethods benchmark defined in groupby.py.

You can also run the benchmark suite using the version of pandas already installed in your current Python environment. This can be useful if you do not have virtualenv or conda, or are using the setup.py develop approach discussed above; for the in-place build you need to set PYTHONPATH, e.g. PYTHONPATH="$PWD/.." asv [remaining arguments]. You can run benchmarks using an existing Python environment by:

asv run -e -E existing

or, to use a specific Python interpreter,:

asv run -e -E existing:python3.6

This will display stderr from the benchmarks, and use your local python that comes from your $PATH.

Information on how to write a benchmark and how to use asv can be found in the asv documentation.

Documenting your code ¶

Changes should be reflected in the release notes located in doc/source/whatsnew/vx.y.z.rst. This file contains an ongoing change log for each release. Add an entry to this file to document your fix, enhancement or (unavoidable) breaking change. Make sure to include the GitHub issue number when adding your entry (using :issue:`1234` where 1234 is the issue/pull request number).

If your code is an enhancement, it is most likely necessary to add usage examples to the existing documentation. This can be done following the section regarding documentation above. Further, to let users know when this feature was added, the versionadded directive is used. The sphinx syntax for that is:

.. versionadded:: 0.21.0

This will put the text New in version 0.21.0 wherever you put the sphinx directive. This should also be put in the docstring when adding a new function or method (example) or a new keyword argument (example).

Contributing your changes to pandas¶

Committing your code ¶

Keep style fixes to a separate commit to make your pull request more readable.

Once you’ve made changes, you can see them by typing:

git status

If you have created a new file, it is not being tracked by git. Add it by typing:

git add path/to/file-to-be-added.py

Doing ‘git status’ again should give something like:

# On branch shiny-new-feature
#
#       modified:   /relative/path/to/file-you-added.py
#

Finally, commit your changes to your local repository with an explanatory message. Pandas uses a convention for commit message prefixes and layout. Here are some common prefixes along with general guidelines for when to use them:

ENH: Enhancement, new functionality
BUG: Bug fix
DOC: Additions/updates to documentation
TST: Additions/updates to tests
BLD: Updates to the build process/scripts
PERF: Performance improvement
TYP: Type annotations
CLN: Code cleanup

The following defines how a commit message should be structured. Please reference the relevant GitHub issues in your commit message using GH1234 or #1234. Either style is fine, but the former is generally preferred:

a subject line with < 80 chars.
One blank line.
Optionally, a commit message body.

Now you can commit your changes in your local repository:

git commit -m

Pushing your changes ¶

When you want your changes to appear publicly on your GitHub page, push your forked feature branch’s commits:

git push origin shiny-new-feature

Here origin is the default name given to your remote repository on GitHub. You can see the remote repositories:

git remote -v

If you added the upstream repository as described above you will see something like:

origin  git@github.com:yourname/pandas.git (fetch)
origin  git@github.com:yourname/pandas.git (push)
upstream        git://github.com/pandas-dev/pandas.git (fetch)
upstream        git://github.com/pandas-dev/pandas.git (push)

Now your code is on GitHub, but it is not yet a part of the pandas project. For that to happen, a pull request needs to be submitted on GitHub.

Review your code ¶

When you’re ready to ask for a code review, file a pull request. Before you do, once again make sure that you have followed all the guidelines outlined in this document regarding code style, tests, performance tests, and documentation. You should also double check your branch changes against the branch it was based on:

Navigate to your repository on GitHub – https://github.com/your-user-name/pandas
Click on Branches
Click on the Compare button for your feature branch
Select the base and compare branches, if necessary. This will be master and shiny-new-feature, respectively.

Finally, make the pull request ¶

If everything looks good, you are ready to make a pull request. A pull request is how code from a local repository becomes available to the GitHub community and can be looked at and eventually merged into the master version. This pull request and its associated changes will eventually be committed to the master branch and available in the next release. To submit a pull request:

Navigate to your repository on GitHub
Click on the Pull Request button
You can then click on Commits and Files Changed to make sure everything looks okay one last time
Write a description of your changes in the Preview Discussion tab
Click Send Pull Request.

This request then goes to the repository maintainers, and they will review the code.

Updating your pull request ¶

Based on the review you get on your pull request, you will probably need to make some changes to the code. In that case, you can make them in your branch, add a new commit to that branch, push it to GitHub, and the pull request will be automatically updated. Pushing them to GitHub again is done by:

git push origin shiny-new-feature

This will automatically update your pull request with the latest code and restart the Continuous Integration tests.

Another reason you might need to update your pull request is to solve conflicts with changes that have been merged into the master branch since you opened your pull request.

To do this, you need to “merge upstream master” in your branch:

git checkout shiny-new-feature
git fetch upstream
git merge upstream/master

If there are no conflicts (or they could be fixed automatically), a file with a default commit message will open, and you can simply save and quit this file.

If there are merge conflicts, you need to solve those conflicts. See for example at https://help.github.com/articles/resolving-a-merge-conflict-using-the-command-line/ for an explanation on how to do this. Once the conflicts are merged and the files where the conflicts were solved are added, you can run git commit to save those fixes.

If you have uncommitted changes at the moment you want to update the branch with master, you will need to stash them prior to updating (see the stash docs). This will effectively store your changes and they can be reapplied after updating.

After the feature branch has been update locally, you can now update your pull request by pushing to the branch on GitHub:

git push origin shiny-new-feature

Delete your merged branch (optional)¶

Once your feature branch is accepted into upstream, you’ll probably want to get rid of the branch. First, merge upstream master into your branch so git knows it is safe to delete your branch:

git fetch upstream
git checkout master
git merge upstream/master

Then you can do:

git branch -d shiny-new-feature

Make sure you use a lower-case -d, or else git won’t warn you if your feature branch has not actually been merged.

The branch will still exist on GitHub, so to delete it there do:

git push origin --delete shiny-new-feature

Tips for a successful Pull Request ¶

If you have made it to the Review your code phase, one of the core contributors may take a look. Please note however that a handful of people are responsible for reviewing all of the contributions, which can often lead to bottlenecks.

To improve the chances of your pull request being reviewed, you should:

Reference an open issue for non-trivial changes to clarify the PR’s purpose
Ensure you have appropriate tests. These should be the first part of any PR
Keep your pull requests as simple as possible. Larger PRs take longer to review
Ensure that CI is in a green state. Reviewers may not even look otherwise
Keep Updating your pull request, either by request or every few days