Datasets:
File size: 8,659 Bytes
94ec51c | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 |
#!/usr/bin/env python3
"""
run_experiments.py
Reproducible experiments for the "Sudoku the Hard Way" p-adic regression solver.
Default behaviour is *fast*: puzzles are produced by carving random solved grids down to a
given clue count (no uniqueness guarantee). This keeps runtimes sane.
Use --unique to enforce unique solutions (slower, but closer to "real" Sudoku).
"""
from __future__ import annotations
import argparse
import csv
import time
from pathlib import Path
import random
import statistics
try:
import matplotlib.pyplot as plt # type: ignore
except ImportError: # pragma: no cover
plt = None
from padic_sudoku_regression import (
generate_unique_puzzle,
random_solved_grid,
grid_to_string,
parse_puzzle,
solve_stepwise_swap,
solve_greedy_descent_swap,
solve_greedy_local_edit_best,
solve_greedy_local_edit_first,
solve_zubarev_local_edit,
solve_zubarev_walk,
pretty,
)
def carve_fast(solved, clues: int, rng: random.Random):
puzzle = solved[:]
positions = list(range(81))
rng.shuffle(positions)
to_remove = 81 - clues
for pos in positions[:to_remove]:
puzzle[pos] = 0
return puzzle
def main() -> None:
ap = argparse.ArgumentParser()
ap.add_argument("--outdir", type=str, default=str(Path(__file__).resolve().parent.parent / "outputs"))
ap.add_argument("--seed", type=int, default=0)
ap.add_argument("--n", type=int, default=12, help="puzzles per clue count")
ap.add_argument("--clues", type=str, default="36,30,26", help="comma-separated clue counts")
ap.add_argument("--max-steps", type=int, default=200000)
ap.add_argument("--restarts", type=int, default=30)
ap.add_argument(
"--method",
type=str,
default="stepwise",
choices=["stepwise", "greedy", "zubarev", "local-best", "local-first", "local-zubarev"],
)
ap.add_argument("--beta0", type=float, default=0.5, help="Zubarev walk: initial beta (inverse temperature).")
ap.add_argument("--beta1", type=float, default=6.0, help="Zubarev walk: final beta (ignored if schedule=constant).")
ap.add_argument("--beta-schedule", type=str, default="linear", choices=["constant", "linear", "exp"])
ap.add_argument("--unique", action="store_true", help="enforce uniqueness (slower)")
args = ap.parse_args()
outdir = Path(args.outdir)
outdir.mkdir(parents=True, exist_ok=True)
clue_counts = [int(x.strip()) for x in args.clues.split(",") if x.strip()]
rng = random.Random(args.seed)
results_rows = []
# We'll capture a trace for the first puzzle in the middle clue count (if available).
trace_target = clue_counts[len(clue_counts)//2]
trace_saved = False
t0 = time.time()
for clues in clue_counts:
for j in range(args.n):
seed = rng.randrange(1_000_000_000)
if args.unique:
puzzle = generate_unique_puzzle(clues=clues, seed=seed)
else:
solved = random_solved_grid(random.Random(seed))
puzzle = carve_fast(solved, clues=clues, rng=random.Random(seed ^ 0xDEADBEEF))
puzzle_str = grid_to_string(puzzle)
record_trace = (not trace_saved) and (clues == trace_target) and (j == 0)
if args.method == "stepwise":
res = solve_stepwise_swap(
puzzle,
seed=seed ^ 0xA5A5A5A5,
max_steps=args.max_steps,
restarts=args.restarts,
record_trace=record_trace,
trace_every=200,
)
elif args.method == "greedy":
res = solve_greedy_descent_swap(
puzzle,
seed=seed ^ 0xA5A5A5A5,
max_steps=args.max_steps,
restarts=args.restarts,
record_trace=record_trace,
trace_every=200,
)
elif args.method == "local-best":
res = solve_greedy_local_edit_best(
puzzle,
seed=seed ^ 0xA5A5A5A5,
max_steps=args.max_steps,
restarts=args.restarts,
record_trace=record_trace,
trace_every=200,
)
elif args.method == "local-first":
res = solve_greedy_local_edit_first(
puzzle,
seed=seed ^ 0xA5A5A5A5,
max_steps=args.max_steps,
restarts=args.restarts,
record_trace=record_trace,
trace_every=200,
)
elif args.method == "local-zubarev":
res = solve_zubarev_local_edit(
puzzle,
seed=seed ^ 0xA5A5A5A5,
max_steps=args.max_steps,
restarts=args.restarts,
beta0=args.beta0,
beta1=args.beta1,
beta_schedule=args.beta_schedule,
record_trace=record_trace,
trace_every=200,
)
else:
res = solve_zubarev_walk(
puzzle,
seed=seed ^ 0xA5A5A5A5,
max_steps=args.max_steps,
restarts=args.restarts,
beta0=args.beta0,
beta1=args.beta1,
beta_schedule=args.beta_schedule,
record_trace=record_trace,
trace_every=200,
)
results_rows.append({
"clues": clues,
"puzzle_seed": seed,
"solve_seed": seed ^ 0xA5A5A5A5,
"unique_enforced": int(args.unique),
"method": args.method,
"objective": res.objective_label,
"puzzle": puzzle_str,
"solved": int(res.solved),
"steps": res.steps,
"restarts_used": res.restarts,
"seconds": res.seconds,
"final_conflicts": res.final_conflicts,
})
if record_trace and res.trace is not None:
# Save trace plot
if plt is None:
print("Note: matplotlib not installed; skipping loss_curve plot.")
else:
fig = plt.figure()
x_values = res.trace_steps if res.trace_steps is not None else [200 * k for k in range(len(res.trace))]
plt.plot(x_values, res.trace)
plt.xlabel("Iteration (approx.)")
plt.ylabel(f"{res.objective_label} conflict pairs")
plt.title(f"Loss trajectory (clues={clues}, seed={seed})")
fig.tight_layout()
png_path = outdir / "loss_curve.png"
pdf_path = outdir / "loss_curve.pdf"
fig.savefig(png_path, dpi=200)
fig.savefig(pdf_path)
plt.close(fig)
# also save the puzzle and (if solved) its solution
(outdir / "trace_puzzle.txt").write_text(pretty(parse_puzzle(puzzle_str)))
if res.solved:
(outdir / "trace_solution.txt").write_text(pretty(res.grid))
trace_saved = True
# Write CSV
csv_path = outdir / "experiment_results.csv"
with csv_path.open("w", newline="") as f:
w = csv.DictWriter(f, fieldnames=list(results_rows[0].keys()))
w.writeheader()
for row in results_rows:
w.writerow(row)
# Summaries
summary_lines = []
summary_lines.append(f"Total runs: {len(results_rows)}")
summary_lines.append(f"Unique enforced: {args.unique}")
for clues in clue_counts:
subset = [r for r in results_rows if r["clues"] == clues]
solved = [r for r in subset if r["solved"] == 1]
summary_lines.append(f"\nClues={clues}: solved {len(solved)}/{len(subset)}")
if solved:
steps = [r["steps"] for r in solved]
secs = [r["seconds"] for r in solved]
summary_lines.append(f" median steps: {int(statistics.median(steps))}")
summary_lines.append(f" median seconds: {statistics.median(secs):.3f}")
(outdir / "summary.txt").write_text("\n".join(summary_lines))
dt = time.time() - t0
print("Wrote", csv_path)
print("Wall time:", f"{dt:.2f}s")
print((outdir / "summary.txt").read_text())
if __name__ == "__main__":
main()
|