Testing#

This page shows where the tests are, what each group of tests checks, and how to run them.

Run the tests#

Run the three checks from the source folder. All three must pass before a pull request is ready:

pip

pytest           # all tests
ruff check .     # code style
mypy src         # types

uv

uv run pytest           # all tests
uv run ruff check .     # code style
uv run mypy src         # types

To make mypy pass, change the code. Keep the mypy settings as they are.

Test folders#

tests/
├── test_*.py              tests per module or feature: session, adapters, data model, export, workflows, ...
├── regression/            comparison tests with eitprocessing and resurfemg
│   └── real_data/         tests on recorded data, not part of the normal test run
├── snapshots/
│   └── workflow_specs/    saved structure of each example workflow file
└── data/                  synthetic recordings for the tests and the examples

Comparison tests#

The tests in tests/regression/ compare M3Resp with the original packages. Each test runs an M3Resp function on synthetic data, and the original function with the same settings. Most tests require exactly equal values (assert_array_equal, or atol=0, rtol=0).

Test file	Compares
`test_eit_equivalence.py`, `test_eit_gap_migration_equivalence.py`	`EITProcessingAdapter` with `eitprocessing`
`test_emg_equivalence.py`	`ReSurfEMGAdapter` with `resurfemg`
`test_processing_filter_equivalence.py`, `test_processing_metric_equivalence.py`, `test_processing_peak_interval_equivalence.py`, `test_processing_quality_equivalence.py`, `test_processing_window_equivalence.py`	The copies in `m3resp.processing` with their originals
`test_ecg_peak_detection_ground_truth.py`	The ECG peak detection with the R peaks of a simulated ECG signal, found by `neurokit2`

If one of these tests fails, the M3Resp result differs from the original result. Look at the M3Resp function first, before you change the test. See Adapters.

Tests on recorded data#

The tests in tests/regression/real_data/ run M3Resp on full recordings: dataset 3 of the recordings of 27 July 2026 (TestPS3). This dataset has Biopac EMG and airway pressure, FluxMed flow and Dräger EIT. The tests use one minute with the EIT on (37 to 97 s on the Biopac clock) and one minute with the EIT off (140 to 200 s). For each minute, the tests put the FluxMed and EIT recordings on the Biopac clock. They use the cross-correlation of the airway pressure, which all three devices record.

The recordings are kept outside the repository. The tests read them from the folder in the environment variable M3RESP_REAL_DATA, with data/source/Test_PS_27072026 as the default. They check the SHA-256 checksum of each file against datasets.yaml, and skip if a file is missing. The tests have the marker real_data, and the normal test run leaves them out. To run them:

pip

pytest -m real_data

uv

uv run pytest -m real_data

The tests make three kinds of check:

Same as the original packages: each M3Resp step gives exactly the result of eitprocessing or resurfemg, called directly on the same data.
Same as the last time: each number equals the stored value in reference_values.json. Counts and sample numbers must be exactly equal. Other numbers can have a relative difference of at most 1e-9.
Physiology: the heart rate of the EIT against the ECG in the EMG, the breathing rates and breath starts against the flow, and the EMG filter chains with a notch filter against each other while the EIT is on.

If you change the code to change the results, write the new reference values with scripts/update_real_data_references.py. Then look at the changes in reference_values.json, which show the numbers that changed.

Tests without the optional packages#

tests/test_optional_dependency_absence.py checks that you can import m3resp, list the steps and check a workflow file without eitprocessing and resurfemg. The test hides the two packages in a Python environment that has them.

The job optional-deps in .github/workflows/tests.yml installs M3Resp four times in new environments: with no extras, and with .[eit], .[emg] and .[all]. Each time, it runs python -m m3resp steps and scripts/check_optional_dependency_isolation.py. The script checks that the installed packages match the extra, and that each step shows the right status: "available" or "missing_optional_dependency".

Snapshots of the example workflows#

tests/test_workflow_spec_baseline.py and tests/test_example_specs.py load and check each example workflow file in examples/. They compare its structure with the saved JSON files in tests/snapshots/workflow_specs/: the order of the steps, the in, with and out of each step, and the results folder. So a change in how M3Resp reads workflow files shows up as a failed test.

Make a new snapshot only after you checked that the change is correct, and that no other part of the structure changed.

Test data#

The recordings in tests/data are synthetic. The tests, the tutorials and the example workflows use these files. m3resp.synthetic makes synthetic EIT, EMG and ventilator recordings. The m3resp_multimodal_1_* files come from the settings in src/m3resp/synthetic/synthetic_generator_config.yaml.

Warning

Do not add patient recordings to the repository. Use synthetic recordings for tests and examples.

Continue with Coding guidelines.