Developer Start Guide#


Looking for an idea for a first contribution? Check the GitHub issues.

We recommend creating an issue to discuss proposed changes before making them. This is a good way to make sure that proposed changes will be accepted.


See Community for a list of PyTensor resources.

The Theano Google group is also relevant to (early) PyTensor versions: theano-dev.

Requirements for Quality Contributions#

The following are requirements for a quality pull request (PR)/contribution:

  • All code should be accompanied by quality unit tests that provide complete coverage of the features added.

  • There is an informative high-level description of the changes, or a reference to an issue describing the changes.

  • The description and/or commit messages reference any relevant GitHub issues.

  • pre-commit is installed and set up.

  • The commit messages follow these guidelines.

  • The commits correspond to relevant logical changes, and there are no commits that fix changes introduced by other commits in the same branch/BR.

  • There are tests, implemented within the pytest framework, covering the changes introduced by the PR.

  • Type hints are added where appropriate.

Don’t worry, your PR doesn’t need to be in perfect order to submit it. As development progresses and/or reviewers request changes, you can always rewrite the history of your feature/PR branches.

If your PR is an ongoing effort and you would like to involve us in the process, simply make it a draft PR.

When you submit a PR, your changes will automatically be tested via our continuous integration (CI). Just because the tests run automatically does not mean you shouldn’t run them yourself to make sure everything is all right. You can run only the portion you are modifying to go faster and have CI make sure there are no broader problems.

To run the test suite with the default options, see How to test that PyTensor works properly.

Documentation and docstrings#

A Docstring Example#

Here is an example on how to add a docstring to a class.

from pytensor.graph.basic import Variable
from pytensor.graph.op import Op

class DoubleOp(Op):
    """Double each element of a tensor.

    this is a test note

    See Also
    `Elemwise`: This functionality is already available; just execute
    ``x * 2`` with ``x`` being an PyTensor variable.

    def make_node(self, x: Variable):
        """Construct an `Apply` node for this `Op`.

            Input tensor


Installation and configuration#

To submit PRs, create an account on GitHub and fork PyTensor.

This will create your own clone of the PyTensor project on GitHub’s servers. It is customary to assign this Git remote the name “origin”, and the official PyTensor repository the name “upstream”.

Create a local copy#

Clone your fork locally with

git clone

For this URL to work, you must set your public SSH keys inside your GitHub account setting.

From your local repository, your fork on GitHub will be called “origin” by default.

Next, create a remote entry for the original (i.e. upstream) PyTensor repository with the following:

git remote add upstream git://


You can choose a name other than “upstream” to reference the official PyTensor repository.

Setting up the your local development environment#

You will need to create a virtual environment and install the project requirements within it.

The recommended approach is to install conda and create a virtual environment in the project directory:

conda env create -f environment.yml
conda activate pytensor-dev

Next, pre-commit needs to be configured so that the linting and code quality checks are performed before each commit:

pre-commit install

The virtual environment will need to be activated in any environment (e.g. shells, IDEs, etc.) that plans to run the PyTensor tests or add commits to the project repository.

You can now test your environment/code by running pytest in the project’s root directory. See How to test that PyTensor works properly for more information about testing.

For a general guide on how to provide open source contributions see here. For a good overview of the development workflow (e.g. relevant git commands) see the NumPy development guide.

Contributing to the documentation#

The documentation build dependencies have also been included in the virtual environment you created. You can also create a separate virtual environment just for the documentation using the environment.yml file located inside the doc folder.

conda env create -f doc/environment.yml
conda activate pytensor-docs

You can now build the documentation from the root of the project with:

python -m sphinx -b html ./doc ./html

Afterward, you can go to html/index.html and navigate the changes in a browser. One way to do this is to go to the html directory and run:

python -m http.server

Do not commit the `html` directory. The documentation is built automatically. For more documentation customizations such as different formats e.g., PDF, refer to the Sphinx documentation.

Other tools that might help#

  • cProfile: Time profiler that work at function level

  • line_profiler: Line-by-line profiler

  • memory_profiler: A memory profiler

  • runsnake: GUI for cProfile (time profiler) and Meliae (memory profiler)

  • Guppy: Supports object and heap memory sizing, profiling, and debugging

  • hub: A tool that adds GitHub commands to the git command line

  • git pull-requests: Another command line tool for git/GitHub