Pydra FSL Tasks

This repository aims to be the canonical set of Pydra tasks for incorporating FSL tools into a Pydra workflow.

Part of this effort is to establish a (mostly) declarative language for describing tasks that potentially have intricate rules for determining the availability and names from the choice of inputs. See Converter for this aspect of the repository.

Installation

pip install /path/to/pydra-fsl/

Installation for developers

pip install -e /path/to/pydra-fsl/[dev]

Converter

FSLConverter class (from tools/converter.py) requires three parts of information:

Nipype spec: converter loads nipype interface and reads _cmd, input_spec and output_spec
yml file with additional spec: specs/fsl_{module_name}_params.yml contains additional spec that are written based on additional functions from nipype (including list_outputs), each interface can have the following fields:
- inputs_metadata: additional metadata for fields from input_spec (it will be included in metadata in pydra spec), e.g., used in specs/fsl_preprocess_params.yml for FAST to set default value for number_classes (it's not part of nipype's spec, but it's set in list_output, see here)
- output_requirements: providing required fields for the output to be created, taken from list_output structure (e.g. requirements for tissue_class_files here); it's a part of the requires field in metadata in pydra spec (e.g. here)
- output_templates: providing template to create the output file name, taken from list_output structure (e.g., here); it is set as output_file_template in metadata (e.g. here)
- output_callables: providing function name that should be used to gather output, based on the list_output structure and used only for FAST; it is set as callable in metadata (e.g. here)
- tests_inputs, tests_outputs: specification for tests, the fields should have the same length and each element should contain the input fields values and list of the expected output fields names
- doctests: specification for doctest, should include values for input fields and the expected cmdline
python file with functions used as callables to gather the outputs: specs/callables.py should contain all the functions from output_callables; the source code of the functions is read and written again in the pydra interface file

How to use the convert

The converter can be used by running:

python tools/converter.py --interface_name <name of teh interface> --module_name <module_name>

The pydra task will be created and saved in pydra/tasks/fsl/{module_name}/{interface_name}.py. Note, that the spec file has to be present for the specific module name in order to run the converter. If no interface_name is provided, the default value all will be used and the converter will be run for all interfaces from the spec file.

Tests are written based on the fields from the yml file: tests_inputs and tests_outputs (the lengths should be the same). One test, test_specs_* checks only the correctness of the spec based on the test_inputs/outputs pairs, i.e. predicts which output fields should be created based on the list of the input fields. The second test, test_run_* should run the interfaces (TODO: this is temporary, should be removed from the final repo). Tests can be run using pytest:

pytest -vs pydra/tasks/fsl/{module_name}/tests

Interface progress

Below is a list of all planned interfaces, with completed interfaces checked. The list was copied from the nipype documentation at https://nipype.readthedocs.io/en/latest/api/generated/nipype.interfaces.fsl.html.

astewartau/pydra-fsl

Pydra FSL Tasks

Installation

Installation for developers

Converter

How to use the convert

Interface progress

Preprocess

AROMA

DTI

EPI

FIX

Utils

POSSUM

Model

Maths