itt_solver/gates.py

from .solver_core import tile_transform
import numpy as np

def gate_boundary_respect(phi_candidate, phi_in, phi_target, boundary_mask=None):
    """Gate A: candidate must not change locked boundary cells.
    If boundary_mask is None, infer from phi_in != 0.
    If mask shape differs from candidate, resize by tiling.
    """
    if boundary_mask is None:
        boundary_mask = (phi_in != 0)

    # If mask shape differs, resize it to candidate shape using tiling
    if boundary_mask.shape != phi_candidate.shape:
        # convert to int for tile_transform then back to bool
        try:
            resized = tile_transform(boundary_mask.astype(int), phi_candidate.shape).astype(bool)
        except Exception:
            # fallback: broadcast along axes if possible
            resized = np.broadcast_to(boundary_mask, phi_candidate.shape)
    else:
        resized = boundary_mask

    # If any boundary cell in phi_target differs from candidate where resized True, reject
    changed = np.any((phi_candidate[resized] != phi_target[resized]))
    return not changed


def gate_sigma_localization(phi_candidate, phi_target, max_fraction=0.5):
    """Gate B: ensure residue is localized. Compute fraction of cells that differ; must be <= max_fraction."""
    diff = (phi_candidate != phi_target)
    frac = np.sum(diff) / diff.size
    return frac <= max_fraction

def gate_quantization(phi_candidate, allowed_symbols=None, tol=1e-6):
    """Gate C: candidate values must quantize to allowed_symbols (integers 0-9 by default).
    Allow small numerical tolerance.
    """
    if allowed_symbols is None:
        allowed_symbols = list(range(10))
    # Round candidate to nearest integer and check membership
    rounded = np.rint(phi_candidate).astype(int)
    mask = np.isin(rounded, allowed_symbols)
    return np.all(mask)

def validate_gates(phi_candidate, phi_in, phi_target, boundary_mask=None, max_fraction=0.5, allowed_symbols=None):
    """Run all gates and return a dict of booleans and an overall pass boolean."""
    a = gate_boundary_respect(phi_candidate, phi_in, phi_target, boundary_mask)
    b = gate_sigma_localization(phi_candidate, phi_target, max_fraction=max_fraction)
    c = gate_quantization(phi_candidate, allowed_symbols=allowed_symbols)
    passed = a and b and c
    return {"A_boundary": a, "B_localization": b, "C_quantization": c, "passed": passed}