134 lines
5.9 KiB
Python
134 lines
5.9 KiB
Python
#!/usr/bin/env python3
|
|
"""Tests for density evolution optimizer."""
|
|
|
|
import numpy as np
|
|
import pytest
|
|
import sys
|
|
import os
|
|
|
|
sys.path.insert(0, os.path.dirname(__file__))
|
|
|
|
|
|
class TestDensityEvolution:
|
|
"""Tests for the Monte Carlo DE engine."""
|
|
|
|
def test_de_known_good_converges(self):
|
|
"""DE with original staircase profile at lam_s=10 should converge easily."""
|
|
from density_evolution import run_de, ORIGINAL_STAIRCASE_PROFILE
|
|
np.random.seed(42)
|
|
converged, error_frac = run_de(
|
|
ORIGINAL_STAIRCASE_PROFILE, lam_s=10.0, lam_b=0.1,
|
|
z_pop=10000, max_iter=50
|
|
)
|
|
assert converged, f"DE should converge at lam_s=10, error_frac={error_frac}"
|
|
|
|
def test_de_known_bad_fails(self):
|
|
"""DE at very low lam_s=0.1 should not converge."""
|
|
from density_evolution import run_de, ORIGINAL_STAIRCASE_PROFILE
|
|
np.random.seed(42)
|
|
converged, error_frac = run_de(
|
|
ORIGINAL_STAIRCASE_PROFILE, lam_s=0.1, lam_b=0.1,
|
|
z_pop=10000, max_iter=50
|
|
)
|
|
assert not converged, f"DE should NOT converge at lam_s=0.1, error_frac={error_frac}"
|
|
|
|
def test_de_population_shape(self):
|
|
"""Verify belief arrays have correct shapes after one step."""
|
|
from density_evolution import de_channel_init, density_evolution_step
|
|
np.random.seed(42)
|
|
n_base = 8
|
|
m_base = 7
|
|
z_pop = 1000
|
|
|
|
# Original staircase H_base profile
|
|
from density_evolution import ORIGINAL_STAIRCASE_PROFILE
|
|
beliefs, msg_memory = de_channel_init(ORIGINAL_STAIRCASE_PROFILE, z_pop, lam_s=5.0, lam_b=0.1)
|
|
|
|
# beliefs should be (n_base, z_pop)
|
|
assert beliefs.shape == (n_base, z_pop), f"Expected ({n_base}, {z_pop}), got {beliefs.shape}"
|
|
|
|
# Run one step
|
|
beliefs = density_evolution_step(beliefs, msg_memory, ORIGINAL_STAIRCASE_PROFILE, z_pop)
|
|
assert beliefs.shape == (n_base, z_pop), f"Shape changed after step: {beliefs.shape}"
|
|
|
|
|
|
class TestThresholdComputation:
|
|
"""Tests for threshold binary search."""
|
|
|
|
def test_threshold_original_staircase(self):
|
|
"""Threshold for original staircase [7,2,2,2,2,2,2,1] should be ~3-6 photons."""
|
|
from density_evolution import compute_threshold_for_profile
|
|
np.random.seed(42)
|
|
threshold = compute_threshold_for_profile(
|
|
[7, 2, 2, 2, 2, 2, 2, 1], m_base=7, lam_b=0.1,
|
|
z_pop=10000, tol=0.5
|
|
)
|
|
assert 2.0 < threshold < 8.0, f"Expected threshold ~3-6, got {threshold}"
|
|
|
|
def test_threshold_peg_ring(self):
|
|
"""PEG ring [7,3,3,3,2,2,2,2] should have lower or equal threshold than original."""
|
|
from density_evolution import compute_threshold_for_profile
|
|
np.random.seed(42)
|
|
thresh_orig = compute_threshold_for_profile(
|
|
[7, 2, 2, 2, 2, 2, 2, 1], m_base=7, lam_b=0.1,
|
|
z_pop=15000, tol=0.25
|
|
)
|
|
np.random.seed(123)
|
|
thresh_peg = compute_threshold_for_profile(
|
|
[7, 3, 3, 3, 2, 2, 2, 2], m_base=7, lam_b=0.1,
|
|
z_pop=15000, tol=0.25
|
|
)
|
|
assert thresh_peg <= thresh_orig, (
|
|
f"PEG threshold {thresh_peg} should be <= original {thresh_orig}"
|
|
)
|
|
|
|
def test_profile_to_hbase(self):
|
|
"""build_de_profile should produce valid profile with correct column degrees."""
|
|
from density_evolution import build_de_profile
|
|
profile = build_de_profile([7, 3, 2, 2, 2, 2, 2, 2], m_base=7)
|
|
assert profile['n_base'] == 8
|
|
assert profile['m_base'] == 7
|
|
assert profile['vn_degrees'] == [7, 3, 2, 2, 2, 2, 2, 2]
|
|
# Every row should have at least 2 connections
|
|
for r, conns in enumerate(profile['connections']):
|
|
assert len(conns) >= 2, f"Row {r} has only {len(conns)} connections"
|
|
|
|
|
|
class TestDegreeDistributionOptimizer:
|
|
"""Tests for the exhaustive search optimizer."""
|
|
|
|
def test_enumerate_candidates(self):
|
|
"""Enumeration should produce 3^7 = 2187 candidates."""
|
|
from density_evolution import enumerate_vn_candidates
|
|
candidates = enumerate_vn_candidates(m_base=7)
|
|
assert len(candidates) == 3**7, f"Expected 2187, got {len(candidates)}"
|
|
# Each candidate should have 8 elements (info col + 7 parity)
|
|
for c in candidates:
|
|
assert len(c) == 8
|
|
assert c[0] == 7 # info column always degree 7
|
|
|
|
def test_filter_removes_invalid(self):
|
|
"""Filter should keep valid distributions and remove truly invalid ones."""
|
|
from density_evolution import filter_by_row_degree
|
|
# All-dv=2 parity: parity_edges=14, dc_avg=3 -> valid for [3,6]
|
|
all_2 = [7, 2, 2, 2, 2, 2, 2, 2]
|
|
assert filter_by_row_degree([all_2], m_base=7, dc_min=3, dc_max=6) == [all_2]
|
|
# All-dv=4 parity: parity_edges=28, dc_avg=5 -> valid for [3,6]
|
|
all_4 = [7, 4, 4, 4, 4, 4, 4, 4]
|
|
assert filter_by_row_degree([all_4], m_base=7, dc_min=3, dc_max=6) == [all_4]
|
|
# A hypothetical all-dv=1 parity: parity_edges=7, total=14, avg dc=2 < 3 -> invalid
|
|
all_1 = [7, 1, 1, 1, 1, 1, 1, 1]
|
|
assert filter_by_row_degree([all_1], m_base=7, dc_min=3, dc_max=6) == []
|
|
# With tighter constraints (dc_min=4), all-dv=2 should be removed
|
|
assert filter_by_row_degree([all_2], m_base=7, dc_min=4, dc_max=6) == []
|
|
|
|
def test_optimizer_finds_better_than_original(self):
|
|
"""Optimizer should find a distribution with threshold <= original staircase."""
|
|
from density_evolution import optimize_degree_distribution, compute_threshold_for_profile
|
|
np.random.seed(42)
|
|
results = optimize_degree_distribution(m_base=7, lam_b=0.1, top_k=5, z_pop_coarse=5000, z_pop_fine=10000, tol=0.5)
|
|
assert len(results) > 0, "Optimizer should return at least one result"
|
|
best_degrees, best_threshold = results[0]
|
|
# Original staircase threshold is ~3-5 photons
|
|
assert best_threshold < 6.0, f"Best threshold {best_threshold} should be < 6.0"
|