File size: 11,256 Bytes
1fc8dae | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 | from __future__ import annotations
import csv
import json
from collections import defaultdict
from pathlib import Path
import matplotlib
matplotlib.use("Agg")
import matplotlib.pyplot as plt
import numpy as np
ROOT = Path(__file__).resolve().parents[1]
LEARNED = ROOT / "artifacts" / "phase13" / "learned_results_v2"
HEURISTICS = ROOT / "artifacts" / "phase13" / "heuristics"
MANIFEST = ROOT / "artifacts" / "phase10" / "final" / "selected_manifest.jsonl"
OUTPUT = ROOT / "artifacts" / "phase13" / "analysis"
REPORT = ROOT / "reports" / "phase13_statistical_analysis.json"
BOOTSTRAP_REPS = 10_000
BOOTSTRAP_SEED = 130524
ALPHA = 0.05
LEARNED_METHODS = [
"hutace_full",
"ariadne_graph_attention_ppo_adapted",
"cnn_ppo",
"plain_unet_ppo",
]
HEURISTIC_METHODS = [
"random_local_waypoint",
"nearest_frontier",
"expected_gain_per_estimated_energy",
]
ORACLE = "full_map_greedy_oracle_diagnostic"
METRICS = {
"final_coverage": ("coverage", "higher"),
"coverage_energy_auc": ("coverage_auc", "higher"),
"coverage_at_320": ("coverage_at_320", "higher"),
"energy_to_0_50_censored": ("energy_to_0_50", "lower"),
}
def read_jsonl(path: Path) -> list[dict]:
with path.open("r", encoding="utf-8") as handle:
return [json.loads(line) for line in handle if line.strip()]
def write_csv(path: Path, rows: list[dict]) -> None:
if not rows:
raise RuntimeError(f"Refusing to write empty table: {path}")
with path.open("w", newline="", encoding="utf-8") as handle:
writer = csv.DictWriter(handle, fieldnames=list(rows[0]))
writer.writeheader()
writer.writerows(rows)
def load_results():
learned_rows: dict[str, list[tuple[int, list[dict]]]] = defaultdict(list)
learned_curves: dict[str, list[np.ndarray]] = defaultdict(list)
for directory in sorted(path for path in LEARNED.iterdir() if path.is_dir()):
summary = json.loads((directory / "summary.json").read_text(encoding="utf-8"))
method = summary["model_id"]
learned_rows[method].append((int(summary["seed"]), read_jsonl(directory / "test_rows.jsonl")))
learned_curves[method].append(np.load(directory / "test_curves.npz")["curves"])
for method in learned_rows:
learned_rows[method].sort(key=lambda item: item[0])
heuristic_rows = {
method: read_jsonl(HEURISTICS / f"{method}_test_rows.jsonl")
for method in HEURISTIC_METHODS + [ORACLE]
}
heuristic_curves = {
method: np.load(HEURISTICS / f"{method}_test_curves.npz")["curves"]
for method in HEURISTIC_METHODS + [ORACLE]
}
return learned_rows, learned_curves, heuristic_rows, heuristic_curves
def validate_alignment(learned_rows, heuristic_rows) -> list[tuple[str, int]]:
if set(learned_rows) != set(LEARNED_METHODS):
raise RuntimeError(f"Learned method mismatch: {set(learned_rows)}")
reference = None
for method in LEARNED_METHODS:
if len(learned_rows[method]) != 10:
raise RuntimeError(f"Expected 10 seeds for {method}")
for _, rows in learned_rows[method]:
keys = [(row["terrain_id"], int(row["start_seed"])) for row in rows]
if len(keys) != 256 or len(set(keys)) != 256:
raise RuntimeError(f"Episode-key failure for {method}")
reference = keys if reference is None else reference
if keys != reference:
raise RuntimeError(f"Episode alignment failure for {method}")
for method, rows in heuristic_rows.items():
keys = [(row["terrain_id"], int(row["start_seed"])) for row in rows]
if keys != reference:
raise RuntimeError(f"Episode alignment failure for {method}")
return reference
def holm_adjust(p_values: list[float]) -> list[float]:
order = np.argsort(p_values)
adjusted = np.empty(len(p_values), dtype=float)
running = 0.0
total = len(p_values)
for rank, index in enumerate(order):
value = min(1.0, (total - rank) * p_values[index])
running = max(running, value)
adjusted[index] = running
return adjusted.tolist()
def main() -> None:
OUTPUT.mkdir(parents=True, exist_ok=True)
learned_rows, learned_curves, heuristic_rows, heuristic_curves = load_results()
keys = validate_alignment(learned_rows, heuristic_rows)
metadata = {
row["candidate_id"]: row
for row in read_jsonl(MANIFEST)
if row["split"] == "test"
}
strata: dict[tuple[str, str], list[int]] = defaultdict(list)
for index, (terrain_id, _) in enumerate(keys):
row = metadata[terrain_id]
strata[(row["continent"], row["stratum"])].append(index)
arrays: dict[str, dict[str, np.ndarray]] = {}
seeds: dict[str, list[int]] = {}
for method, runs in learned_rows.items():
seeds[method] = [seed for seed, _ in runs]
arrays[method] = {
metric: np.asarray([[row[field] for row in rows] for _, rows in runs], dtype=float)
for metric, (field, _) in METRICS.items()
}
for method, rows in heuristic_rows.items():
arrays[method] = {
metric: np.asarray([[row[field] for row in rows]], dtype=float)
for metric, (field, _) in METRICS.items()
}
if any(seeds[method] != seeds["hutace_full"] for method in LEARNED_METHODS):
raise RuntimeError("Learned methods do not share paired training seeds")
overall = []
for method in LEARNED_METHODS + HEURISTIC_METHODS + [ORACLE]:
for metric in METRICS:
per_seed = arrays[method][metric].mean(axis=1)
overall.append({
"method": method,
"method_type": "learned" if method in LEARNED_METHODS else ("oracle_diagnostic" if method == ORACLE else "heuristic"),
"metric": metric,
"mean": float(per_seed.mean()),
"training_seed_sd": float(per_seed.std(ddof=1)) if len(per_seed) > 1 else "",
"training_seeds": len(per_seed) if method in LEARNED_METHODS else 0,
"test_episodes": 256,
})
write_csv(OUTPUT / "overall_metrics.csv", overall)
rng = np.random.default_rng(BOOTSTRAP_SEED)
terrain_samples = np.empty((BOOTSTRAP_REPS, len(keys)), dtype=np.int32)
cursor = 0
for group in sorted(strata):
indices = np.asarray(strata[group], dtype=np.int32)
width = len(indices)
terrain_samples[:, cursor:cursor + width] = rng.choice(
indices, size=(BOOTSTRAP_REPS, width), replace=True
)
cursor += width
seed_samples = rng.integers(0, 10, size=(BOOTSTRAP_REPS, 10), dtype=np.int16)
comparisons = []
raw_p_values = []
for baseline in LEARNED_METHODS[1:] + HEURISTIC_METHODS:
for metric, (_, direction) in METRICS.items():
hut = arrays["hutace_full"][metric]
hut_boot = hut[seed_samples[:, :, None], terrain_samples[:, None, :]].mean(axis=(1, 2))
base = arrays[baseline][metric]
if baseline in LEARNED_METHODS:
base_boot = base[seed_samples[:, :, None], terrain_samples[:, None, :]].mean(axis=(1, 2))
else:
base_boot = base[0, terrain_samples].mean(axis=1)
raw_diff = hut_boot - base_boot
oriented = raw_diff if direction == "higher" else -raw_diff
point_raw = float(hut.mean() - base.mean())
probability = float(np.mean(oriented > 0.0))
p_value = min(1.0, 2.0 * min(np.mean(oriented <= 0.0), np.mean(oriented >= 0.0)))
row = {
"model": "hutace_full",
"baseline": baseline,
"metric": metric,
"preferred_direction": direction,
"hutace_mean": float(hut.mean()),
"baseline_mean": float(base.mean()),
"raw_difference_hutace_minus_baseline": point_raw,
"oriented_improvement": point_raw if direction == "higher" else -point_raw,
"ci95_low_oriented": float(np.quantile(oriented, 0.025)),
"ci95_high_oriented": float(np.quantile(oriented, 0.975)),
"probability_improvement": probability,
"bootstrap_two_sided_p": float(p_value),
}
comparisons.append(row)
raw_p_values.append(float(p_value))
adjusted = holm_adjust(raw_p_values)
for row, p_adjusted in zip(comparisons, adjusted):
row["holm_adjusted_p"] = p_adjusted
row["holm_significant_0_05"] = p_adjusted < ALPHA
write_csv(OUTPUT / "paired_hierarchical_bootstrap_holm.csv", comparisons)
stratum_rows = []
for (continent, stratum), indices in sorted(strata.items()):
for method in LEARNED_METHODS + HEURISTIC_METHODS + [ORACLE]:
for metric in METRICS:
stratum_rows.append({
"continent": continent,
"stratum": stratum,
"episodes": len(indices),
"method": method,
"metric": metric,
"mean": float(arrays[method][metric][:, indices].mean()),
})
write_csv(OUTPUT / "metrics_by_continent_stratum.csv", stratum_rows)
energy_grid = np.linspace(0.0, 640.0, 65)
plt.figure(figsize=(9, 5.5))
for method in LEARNED_METHODS:
curve = np.mean(np.stack(learned_curves[method]), axis=(0, 1))
plt.plot(energy_grid, curve, label=method, linewidth=2.0 if method == "hutace_full" else 1.4)
for method in HEURISTIC_METHODS:
plt.plot(energy_grid, heuristic_curves[method].mean(axis=0), label=method, linestyle="--", linewidth=1.2)
plt.xlabel("Energy consumed")
plt.ylabel("Observed-map coverage")
plt.ylim(0, 1)
plt.grid(alpha=0.25)
plt.legend(fontsize=7, ncol=2)
plt.tight_layout()
plt.savefig(OUTPUT / "coverage_energy_curves.png", dpi=220)
plt.close()
report = {
"schema_version": 1,
"phase": 13,
"status": "completed",
"gate_passed": True,
"test_accessed": True,
"test_episodes": len(keys),
"learned_seeds_per_method": 10,
"bootstrap": {
"repetitions": BOOTSTRAP_REPS,
"seed": BOOTSTRAP_SEED,
"resampling": "paired training seeds plus test episodes within continent-by-terrain-stratum cells",
"confidence_interval": "percentile 95%",
"multiplicity": "Holm family-wise correction across 24 primary comparisons",
},
"primary_baselines": LEARNED_METHODS[1:] + HEURISTIC_METHODS,
"oracle": {"method": ORACLE, "primary_comparison": False, "purpose": "diagnostic upper reference only"},
"outputs": [
"artifacts/phase13/analysis/overall_metrics.csv",
"artifacts/phase13/analysis/paired_hierarchical_bootstrap_holm.csv",
"artifacts/phase13/analysis/metrics_by_continent_stratum.csv",
"artifacts/phase13/analysis/coverage_energy_curves.png",
],
}
REPORT.write_text(json.dumps(report, indent=2) + "\n", encoding="utf-8")
print("PHASE13_STATISTICS_COMPLETED", flush=True)
if __name__ == "__main__":
main()
|