Contributing to Starfish

If you’re reading this section, you’re probably interested in contributing to Starfish. Welcome and thanks for your interest in contributing!

How can I contribute?

Starfish is designed to specify pipeline recipes for image processing. To support this use, the library is composed as a series of pipeline_component modules. The objects that sub-class pipeline_component spawn a command-line interface that should identically track the API of the python library.

A typical starfish run consists of running one or more image processing filter stages, and identifying features through either a spot- or pixel-based approach. The identified features are then decoded into the genes that they correspond to by mapping the fluorescence channel (and optionally hybridization round) using a codebook. Finally, the filtered data are segmented, identifying which cell each feature belongs to.

Creating a new algorithm for an existing pipeline_component

For example, to add a new image filter, one would:

  1. Create a new python file new_filter.py in the starfish/core/image/Filter/ directory.

  2. Find the corresponding AlgorithmBase for your component. For filters, this is FilterAlgorithm, which is found in starfish/core/image/Filter/_base.py. Import that base into new_filter.py, and have your new algorithm subclass it, e.g. create class NewFilter(FilterAlgorithm)

  3. Implement all required methods from the base class.

That’s it! If at any point something gets confusing, it should be possible to look at existing pipeline components of the same category for guidance on implementation.

Reporting bugs

  • Bugs can be contributed as issues in the starfish repository. Please check to make sure there is no existing issue that describes the problem you have identified before adding your bug.

  • When reporting issues please include as much detail as possible about your operating system, starfish version, slicedimage version, and python version. Whenever possible, please also include a brief, self-contained code example that demonstrates the problem, including a full traceback.

Code contributions

  • Don’t break the build: pull requests are expected to pass all automated CI checks. You can run those checks locally by running make all in starfish repository root.

  • All Pull Request comments must be addressed, even after merge.

  • All code must be reviewed by at least 1 other team member.

  • All code must have typed parameters, and when possible, typed return calls (see PEP484). We also encourage rational use of in-line variable annotation when the type of a newly defined object is not clear (see PEP526.

  • All code must be documented according to numpydoc style guidelines.

  • Numpy provides an excellent development workflow that we encourage you to follow when developing features for starfish!

  • All commits must have informative summaries; try to think about what will still make sense looking back on them next year.

  • All pull requests should describe your testing plan.

  • When merging a pull request, squash commits down to the smallest logical number of commits. In cases where a single commit suffices, use the “Squash and Merge” strategy, since it adds the PR number to the commit name. If multiple commits remain, use “Rebase and Merge”.

Notebook contributions

  • All .ipynb files should have a corresponding .py file. Use nbencdec to generate the corresponding .py file.

  • The .py files allow refactor commands in the codebase to find code in the .py files, which is an important to keep the notebooks working as starfish evolves.

Debugging Errors

First, thank you for using Starfish and SpaceTx-Format! Feedback you provide on features and the user experience is critical to making Starfish a successful tool. Because we iterate quickly on this feedback to add new features, things change often, which can result in your code getting out of sync with your data. When that happens, you may observe errors.

Most of the time, you can fix this problem by pulling the most recent version of the code, reinstalling starfish, and restarting your environment. If you’re using starfish with datasets from spaceTx located on our cloudfront distribution, we’re committed to keeping that data up to date. Updated versions of the notebook will reference the correct data version, and copying over the new link should fix any issues.

For example, if a notebook references in-situ sequencing data from August 23rd, and a breaking change occurs on September 26th, it would be necessary to replace the experiment link to point at data that was updated to work post-update:

- http://spacetx.starfish.data.public.s3.amazonaws.com/browse/formatted/20180823/iss_breast/experiment.json
+ http://spacetx.starfish.data.public.s3.amazonaws.com/browse/formatted/20180926/iss_breast/experiment.json

If you’re using your own data with starfish, you may need to re-run your data ingestion workflow based on starfish.experiment.builder.providers.TileFetcher and starfish.experiment.builder.providers.FetchedTile to generate up-to-date versions of spaceTx-format.

Upgrading to a new version

If you’ve installed from pypi, upgrading is as simple as reinstalling starfish.

pip install --upgrade starfish

If you’ve installed our development version to take advantage of new features in real time, you’ll need to fetch changes and reinstall. Assuming you’ve cloned the respository into ./starfish, you can install the newest version as follows:

cd ./starfish
git checkout master
git pull
pip3 install .

Reporting bugs

Bugs can be contributed as issues in the starfish repository. Please check to make sure there is no existing issue that describes the problem you have identified before adding your bug.

When reporting issues please include as much detail as possible about your operating system, starfish version, slicedimage version, and python version. Much of this can be accomplished by sending us the output of pip freeze:

pip freeze > environment.txt

Whenever possible, please also include a brief, self-contained code example that demonstrates the problem, including a full traceback.