---

name: frechet_gaussian_distance description: Compute the Fréchet Gaussian distance used by FID with stable numerical handling.

Frechet Gaussian Distance

Use this skill when a recovery or implementation needs the frechet_distance module from the TTUR/FID paper without rereading the original repository. Do not use it to claim full GAN reproduction unless real image features, dataset statistics, and training evidence are available.

Inputs

• Numeric arrays or JSON files matching the module contract.
• For recovery orchestration, a module-plan target and runtime handoff.

Outputs

• Deterministic JSON-compatible statistics, distances, traces, or recovery checks.
• Diagnostics sufficient to audit reduced/proxy recovery.

Workflow

1. Validate input shapes and finite numeric values.
2. Execute the module-specific deterministic script in scripts/.
3. Preserve diagnostics and command evidence when used in recovery.
4. Treat synthetic activations or toy dynamics as proxy evidence only.

Validation

Run python tests/test_frechet_gaussian_distance.py from this skill directory.

Limitations

This skill is self-contained and does not require the original TTUR repository. It does not download Inception models or train full GANs.