Skip to content

Data

The repository's large data is too big for GitHub. All of it is stored on the Hugging Face dataset zaneblood1/cmb_lensing_data (~23 GB). The dataset card there describes every file format.

Dataset path Size Contents Written by
camb_grid_spline.npz 7.9 GB the 5D CAMB grid (The CAMB Grid) camb_grid_generation/
joint_inference_data/ 512 MB LCDM chains, one folder per run get_lcdm_samples/ (LCDM Sampling on the HPC)
delensed_field_covariance_data/ 683 MB empirical delensed covariances get_delensed_field_covariance/ (Delensed Covariances)
map_joint_phi_noise_data/ 98 MB map_joint's per-mode N_phi measure_phi_noise/get_map_joint_phi_noise/ (MAP Joint Phi Noise)
sampler_phi_noise_data/ 787 MB the sampler's per-mode N_phi measure_phi_noise/get_sampler_phi_noise/ (Sampler Phi Noise)
map_joint_comparison/ 13 GB map_joint timings and reconstructions, Python vs Julia runtime_comparison/

Downloading

Only the CAMB grid is downloaded automatically (Installation). Fetch anything else with the hf command that huggingface_hub installs, one folder at a time:

hf download zaneblood1/cmb_lensing_data --repo-type dataset \
    --include "joint_inference_data/*" --local-dir cmb_lensing_data

Drop the --include to download everything.

Using it

Nothing in the package looks for these folders at a fixed location: every script takes the path of the file or folder it reads.

#chain analysis of one run, with the forecast
python -m cmb_lensing.chain_analysis cmb_lensing_data/joint_inference_data/joint_inference_2026_10_29_2.5uk \
    --nside 128 --theta_pix 2.5 --noise 2.5 --run_fisher \
    --delensed_covariance cmb_lensing_data/delensed_field_covariance_data/delensed_covariance_output_map_joint_2.5uk/delensed_covariance.npz \
    --phi_noise cmb_lensing_data/map_joint_phi_noise_data/map_joint_phi_noise_output_2.5uk/map_joint_phi_noise.npz

#the forecast on its own
python -m cmb_lensing.fisher_forecast --noise 5 \
    --delensed_covariance cmb_lensing_data/delensed_field_covariance_data/delensed_covariance_output_map_joint_5uk/delensed_covariance.npz

See Chain Analysis and Fisher Forecast for the options. The map_joint comparison figures are remade by runtime_comparison/performance_analysis.py, with the BASE_DIR / RESULTS_DIR at its top pointed at the downloaded map_joint_comparison/performance_results folder.