Nipype2Pydra is a command-line tool for semi-automatically porting command interfaces and workflows written from the Nipype workflow engine to its successor Pydra. It has been designed so that any required manual-specifications are stored in separate YAML specification files, which is read ahead of the conversion, so once designed the conversion can be re-run to pick up changes in newer versions of the Nipype code without having to redo the manual steps.
To convert a Nipype interface to a Pydra task run the nipype2pydra task command and pass it the conversion specification YAML file and the root of the package to save the generated module, e.g.:
$ nipype2pydra task ants_registration_registration.yaml /path/to/package/root
This will create a module file under the package root directory based on the output_module field in the specification, e.g:
/path/to/package/root/output/module/path
If that is missing and the nipype interface is in the standard nipype.interfaces package, then it will be stored at pydra.tasks. with the same path end.
While the conversions aim to as automatic as possible, there are structural differences between Nipype and Pydra that require manual specification, and it might be desirable to rename some of the inputs/outputs to more intuitive names. Such specifications are stored in YAML format, such as the ANTs registration conversion specification,
task_name: Registration
nipype_module: nipype.interfaces.ants.registration
output_requirements:
output_warped_image: ["fixed_image", "moving_image", "output_transform_prefix"]
output_templates:
output_warped_image: "{output_transform_prefix}warped"
doctest:
fixed_image: test.nii.gz
moving_image: test.nii.gz
cmdline: >-
antsRegistration --output [ output_, output_warped_image.nii.gz ]
--metric Mattes[ test.nii, test.nii, 1, 32, Random, 0.05 ]
tests_inputs: []
tests_outputs:
- AttributeErrorDetailed description of the different options to go here
Nipype2Pydra can be installed for Python >= 3.7 from PyPI with
$ python3 -m pip install nipype2pydraThis work is licensed under a Creative Commons Attribution 4.0 International License
