Skip to content

Testing

Most tests are A/B comparisons against CMBLensing.jl: for the same inputs, the Python and Julia outputs should agree to numerical precision. The Julia outputs are stored in tests/ground_truth_data/*.npz.

Test Covers
test_covariance_matrices.py, test_simulated_cls.py load_sim's spectra and covariances
test_lensing.py lense_flow
test_logpdf.py, test_gradients.py the logpdf and its gradients
test_wiener_filter.py, test_map_joint.py the Wiener filter and map_joint
test_camb_grid.py the 5D CAMB grid vs direct CAMB (skipped without the grid)

The sampler, the forecast and the measurement modules are not tested.

Running

pytest                          #all tests
pytest tests/test_lensing.py    #one file

Many tests write comparison plots to tests/test_generated_figures/ rather than asserting. Open tests/index.html with a local server (e.g. VS Code Live Server) and look at the diff plots: a test can pass while a plot shows a problem.

Regenerating the Julia data

Needs juliacall and a local CMBLensing.jl checkout:

export CMBLENSING_JL_PATH=/path/to/CMBLensing.jl
python tests/generate_julia_data/generate_all.py              #all of it
python tests/generate_julia_data/generate_lensing.py          #one test's data
pytest --generate                                             #regenerate, then test

conftest.py maps each test_* module to its generate_* module for --generate.