Thanks for helping to build
Cloning the Repository¶
Make a fork of the dask-ml repo and clone the fork
git clone https://github.com/<your-github-username>/dask-ml cd dask-ml
You may want to add
https://github.com/dask/dask-ml as an upstream remote
git remote add upstream https://github.com/dask/dask-ml
Creating an environment¶
We have conda environment YAML files with all the necessary dependencies
conda env create -f ci/environment-3.7.yml --name=dask-ml-dev
to create a conda environment and install all the dependencies.
If you’re using pip, you can view the list of all the required and optional
setup.py (see the
install_requires field for
required dependencies and
extras_require for optional dependencies).
Dask-ML is a pure-python repository. Development installation should be as simple as cloning the repository and running the following in the cloned directory:
python -m pip install -e ".[dev]"
If you have any trouble, please open an issue on the dask-ml issue tracker.
black . flake8 isort -rc dask_ml tests
You may wish to setup a pre-commit hook to run black when you commit changes.
Dask-ml uses py.test for testing. You can run tests from the main dask-ml directory as follows:
Alternatively you may choose to run only a subset of the full test suite. For example to test only the preprocessing submodule we would run tests as follows:
If your Pull Request decreases the lines of code covered, the CI may fail. Sometimes this is OK, and a maintainer will merge it anyway. To check the coverage locally, use
pytest --cov --cov-report=html
You can still use all the usual pytest command-line options in addition to those.
Install the pre commit tool. Then, from the
root of the
dask-ml repository, run
pre-commit install to install a few plugins
like black, isort, and flake8. These tools will automatically be run on each commit. You
can skip the checks with
git commit --no-verify.
For the most part, we follow scikit-learn’s API design. If you’re implementing a new estimator, it will ideally pass scikit-learn’s estimator check.
We have some additional decisions to make in the dask context. Ideally
- All attributes learned during
.fitshould be concrete, i.e. they should not be dask collections.
- To the extent possible, transformers should support
- If possible, transformers should accept a
columnskeyword to limit the transformation to just those columns, while passing through other columns untouched.
inverse_transformshould behave similarly (ignoring other columns) so that
- Methods returning arrays (like
.predict), should return the same type as the input. So if a
dask.arrayis passed in, a
dask.arraywith the same chunks should be returned.
We use numpydoc for our docstrings.
Examples are written as Jupyter notebooks with their output stripped, either manually or using nbstripout. We want examples to be runnable on binder so they should be small, but include instructions for how to scale up to larger problems.
The source for most examples is maintained in the dask-examples repository. Updates should be made there, and they’re automatically included as part of the Dask-ML documentation build process.
When adding an example for new feature that’s only available in master, the
notebook should be first included in Dask-ML repository under
docs/source/examples/. These examples will be moved to
dask/dask-examples as part of the Dask-ML release process.