Developer Notes

Adding a spectral model

One of the most common changes to the underlying fermitools code is to add a new spectral model. To be able to use that model in fermipy will require a few changes, depending on how exactly you would like you use the model.

  1. At a minimum, the model, and default values and bounds for the parameters need to be added to fermipy/data/models.yaml
  2. If you want to be able to use functions that free the source-shape parameters, fit the SED, you will want to modify the norm_parameters and shape_parameters blocks at the top of the fermipy/gtanalysis.py file to include the new spectral model.
  3. If you want to be able to include the spectral model in an xml ‘catalog’ of sources that you use to define an ROI, you will have to update the spectral_pars_from_catalog and get_catalog_dict functions in fermipy/roi_model.py to include the spectral model.
  4. If the spectral model is included in a new source catalog, and you want to support that catalog, see the section below on supporting new catalogs.
  5. If you want to use the spectral to do more complicated things, like vectorizing call to evalute the spectrum because you are using it in sensitivity studies, then you will have to add it the the fermipy/spectrum.py file. That is pretty much expert territory.

Supporting a new catalog

To support a new catalog will require some changes in the fermipy/catalog.py file. In short

  1. Define a class to manage the catalog. This will have to handle converting the parameters in the FITS file to the format that fermipy expects. It should inherit from the Catalog class.
  2. Update the Catalog.create function to have a hook to create a class of the correct type.
  3. For now we are also maintaining the catalog files in the fermipy/data/catalogs area, so the catalog files should be added to that area.

Creating a New Release

The following are steps for creating a new release:

  1. Update the Changelog page (in docs/changelog.rst) with notes for the release and commit those changes.
  2. Update documentation tables by running make_tables.py inside the docs subdirectory and commit any resulting changes to the configuration table files under docs/config.
  3. Checkout master and ensure that you have pulled all commits from origin.
  4. Create the release tag and push it to GitHub.
$ git tag -a XX.YY.ZZ -m ""
$ git push --tags
  1. Upload the release to pypi.
$ python setup.py sdist upload -r pypi
  1. Create a new release on conda-forge by opening a PR on the fermipy-feedstock repo. There is a fork of fermipy-feedstock in the fermipy organization that you can use for this purpose. Edit recipe/meta.yaml by entering the new package version and updating the sha256 hash to the value copied from the pypi download page. Update the package dependencies as necessary in the run section of requirements. Verify that entry_points contains the desired set of command-line scripts. Generally this section should match the contents entry_points in setup.py. Before merging the PR confirm that all tests have successfully passed.