check_padis_repo_equivalence.py 🚧¶

Source: scripts/dev/check_padis_repo_equivalence.py

Warning

This file has not yet received a complete narrative and docstring audit. Its public source-level API is listed automatically below.

Golden-trace checks between LION PaDIS and the local PaDIS checkout.

This script intentionally does not copy PaDIS implementation code into LION. It reads the relevant oracle functions from the sibling PaDIS repository at runtime, patches only hard-coded device allocation so the check can run on CPU, and compares controlled synthetic traces.

LION_ROOT

class ScaledIdentityOp(scale: float = 1.0): Identity test operator with configurable forward/adjoint scaling.

ScaledIdentityOp.domain_shape(self) → tuple[int, ...]: No docstring is available.

ScaledIdentityOp.range_shape(self) → tuple[int, ...]: No docstring is available.

ScaledIdentityOp.forward(self, x: torch.Tensor) → torch.Tensor: No docstring is available.

ScaledIdentityOp.adjoint(self, y: torch.Tensor) → torch.Tensor: No docstring is available.

ScaledIdentityOp.inverse(self, y: torch.Tensor) → torch.Tensor: No docstring is available.

ScaledIdentityOp.A(self, x: torch.Tensor) → torch.Tensor: No docstring is available.

ScaledIdentityOp.AT(self, y: torch.Tensor) → torch.Tensor: No docstring is available.

ScaledIdentityOp.Adagger(self, y: torch.Tensor) → torch.Tensor: No docstring is available.

class CompatibleDenoiser(pad_width: int = 4, patch_size: int = 4): Deterministic denoiser matching the public PaDIS call convention.

CompatibleDenoiser.forward(self, x: torch.Tensor, sigma_or_c_noise: torch.Tensor, *args, **kwargs): No docstring is available.

class TrainableCompatibleDenoiser(pad_width: int = 0, patch_size: int = 8): Minimal trainable LION denoiser used for equivalence checks.

TrainableCompatibleDenoiser.default_parameters(mode='ct') → LIONModelParameter: No docstring is available.

TrainableCompatibleDenoiser.forward(self, x: torch.Tensor, sigma_or_c_noise: torch.Tensor, *args, **kwargs): No docstring is available.

run_check(args: argparse.Namespace) → dict[str, object]: Run public-repository equivalence checks and collect diagnostics.

build_parser() → argparse.ArgumentParser: Construct the equivalence-check command-line parser.

main() → None: Execute equivalence checks and emit a machine-readable report.