Level 0 Walkthrough¶

Level 0 is a noiseless sanity check: a single galaxy with fixed morphology, very low noise, and a known shear. Since there is effectively no noise, MAP estimation is sufficient -- the point estimate should land directly on the true shear values.

Prerequisites¶

pip install -e .

Step 1: Run a single realization¶

Use shine-bias-run to generate data with a known shear and run MAP inference. The default Level 0 config (configs/validation/level0_base.yaml) uses method: map:

shine-bias-run \
  --shine-config configs/validation/level0_base.yaml \
  --g1-true 0.02 \
  --g2-true -0.01 \
  --seed 42 \
  --output-dir results/validation/level0/r0001 \
  --run-id r0001

This produces:

results/validation/level0/r0001/
├── posterior.nc       # ArviZ InferenceData (MAP point estimate)
├── truth.json         # {"g1": 0.02, "g2": -0.01}
└── convergence.json   # Convergence diagnostics (sentinels for MAP)

Step 2: Extract results¶

Scan the output directory and extract shear estimates into a CSV:

shine-bias-extract \
  --input-dir results/validation/level0 \
  --output results/validation/level0/summary.csv

The CSV contains one row per realization with columns for true shear, estimated shear (mean, median, std), and convergence diagnostics.

Step 3: Compute statistics¶

Compute bias and check acceptance criteria:

shine-bias-stats \
  --input results/validation/level0/summary.csv \
  --output-dir results/validation/level0/stats \
  --level level_0 \
  --posterior-dir results/validation/level0

This produces:

stats/bias_results.json -- bias values and overall pass/fail
stats/plots/ -- diagnostic plots (MAP estimate vs truth)

Acceptance Criteria¶

Level 0 checks three conditions:

Criterion	Threshold	Meaning
Posterior width	\(\sigma < 0.01\)	Posterior should be tight (0 for MAP)
Offset from truth	\(< 1\sigma\)	Estimate should be near truth
Multiplicative bias	\(\\|m\\| < 0.01\)	Less than 1% bias

Inspecting Results¶

Load the posterior in Python for custom analysis:

import arviz as az

idata = az.from_netcdf("results/validation/level0/r0001/posterior.nc")

# Check the inference method
print(idata.posterior.attrs.get("inference_method"))  # "map"

# Point estimates
print(f"g1 = {float(idata.posterior.g1.values.flatten()[0]):.6f}")
print(f"g2 = {float(idata.posterior.g2.values.flatten()[0]):.6f}")

Running Multiple Realizations¶

For more robust validation, loop over seeds:

for seed in 42 43 44 45 46; do
  shine-bias-run \
    --shine-config configs/validation/level0_base.yaml \
    --g1-true 0.02 --g2-true -0.01 \
    --seed $seed \
    --output-dir results/validation/level0/r_s${seed} \
    --run-id r_s${seed}
done

shine-bias-extract \
  --input-dir results/validation/level0 \
  --output results/validation/level0/summary.csv

Or use the batched approach for GPU efficiency -- see GPU-Batched Inference.