Spaces:
Runtime error
Runtime error
File size: 12,085 Bytes
f36b499 | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 | """Trajectory analytics for OpenRange training runs.
Reads JSONL trajectory files (output of ``TrajectoryLogger.export_jsonl``)
and computes summary statistics, per-vuln-class breakdowns, and comparison
reports between runs.
Usage::
python -m open_range.training.analytics trajectories.jsonl
python -m open_range.training.analytics run1.jsonl run2.jsonl --compare
"""
from __future__ import annotations
import argparse
import json
import sys
from pathlib import Path
from typing import Any
class TrajectoryAnalyzer:
"""Analyze JSONL trajectory files produced by TrajectoryLogger.
Each line in a JSONL file is expected to have at minimum::
{
"episode_id": str,
"role": "red" | "blue",
"reward": float,
"outcome": str,
"tier": int,
"messages": [...],
}
Additional fields (snapshot_id, vuln_class, etc.) are used when present.
"""
def __init__(self) -> None:
self._records: list[dict[str, Any]] = []
@property
def records(self) -> list[dict[str, Any]]:
"""All loaded JSONL records."""
return list(self._records)
def load(self, path: str | Path) -> int:
"""Load one or more JSONL files.
Can be called multiple times to accumulate records from
multiple files.
Args:
path: Path to a JSONL file.
Returns:
Number of records loaded from this file.
"""
path = Path(path)
count = 0
with open(path) as f:
for line in f:
line = line.strip()
if not line:
continue
record = json.loads(line)
self._records.append(record)
count += 1
return count
def summary(self) -> dict[str, Any]:
"""Compute summary statistics across all loaded records.
Returns:
Dict with:
- total_episodes: number of unique episode IDs
- total_records: number of JSONL records loaded
- outcomes: dict mapping outcome string to count
- avg_reward: mean reward across all records
- avg_steps: mean step count (from message pairs)
- per_role: dict mapping role to {count, avg_reward, outcomes}
"""
if not self._records:
return {
"total_episodes": 0,
"total_records": 0,
"outcomes": {},
"avg_reward": 0.0,
"avg_steps": 0.0,
"per_role": {},
}
episode_ids = {r.get("episode_id", "") for r in self._records}
# Outcome counts (deduplicated by episode_id)
outcomes: dict[str, int] = {}
seen_episodes: set[str] = set()
for r in self._records:
eid = r.get("episode_id", "")
outcome = r.get("outcome", "unknown")
if eid not in seen_episodes:
outcomes[outcome] = outcomes.get(outcome, 0) + 1
seen_episodes.add(eid)
# Rewards
rewards = [r.get("reward", 0.0) for r in self._records]
avg_reward = sum(rewards) / len(rewards) if rewards else 0.0
# Steps (count assistant messages as steps)
steps_list: list[int] = []
for r in self._records:
messages = r.get("messages", [])
n_steps = sum(1 for m in messages if m.get("role") == "assistant")
steps_list.append(n_steps)
avg_steps = sum(steps_list) / len(steps_list) if steps_list else 0.0
# Per-role stats
per_role: dict[str, dict[str, Any]] = {}
for r in self._records:
role = r.get("role", "unknown")
if role not in per_role:
per_role[role] = {"count": 0, "total_reward": 0.0, "outcomes": {}}
per_role[role]["count"] += 1
per_role[role]["total_reward"] += r.get("reward", 0.0)
outcome = r.get("outcome", "unknown")
per_role[role]["outcomes"][outcome] = (
per_role[role]["outcomes"].get(outcome, 0) + 1
)
for role_data in per_role.values():
count = role_data["count"]
role_data["avg_reward"] = (
role_data["total_reward"] / count if count > 0 else 0.0
)
return {
"total_episodes": len(episode_ids),
"total_records": len(self._records),
"outcomes": outcomes,
"avg_reward": round(avg_reward, 4),
"avg_steps": round(avg_steps, 2),
"per_role": per_role,
}
def by_vuln_class(self) -> dict[str, dict[str, Any]]:
"""Break down solve rates by vulnerability class.
Looks for ``vuln_class`` or ``vuln_classes`` field in records.
For records with ``vuln_classes`` (list), each class is counted
independently.
Returns:
Dict mapping vuln class to:
- attempts: number of episodes
- solves: number of episodes with outcome containing 'win' or 'captured'
- solve_rate: solves / attempts
"""
vuln_stats: dict[str, dict[str, int]] = {}
for r in self._records:
# Only count red records for solve rate
if r.get("role") != "red":
continue
classes: list[str] = []
if "vuln_class" in r:
classes = [r["vuln_class"]]
elif "vuln_classes" in r:
classes = r["vuln_classes"] if isinstance(r["vuln_classes"], list) else [r["vuln_classes"]]
else:
continue
outcome = r.get("outcome", "")
solved = "win" in outcome or "captured" in outcome
for vc in classes:
if vc not in vuln_stats:
vuln_stats[vc] = {"attempts": 0, "solves": 0}
vuln_stats[vc]["attempts"] += 1
if solved:
vuln_stats[vc]["solves"] += 1
result: dict[str, dict[str, Any]] = {}
for vc, stats in sorted(vuln_stats.items()):
result[vc] = {
"attempts": stats["attempts"],
"solves": stats["solves"],
"solve_rate": (
round(stats["solves"] / stats["attempts"], 4)
if stats["attempts"] > 0
else 0.0
),
}
return result
def compare(self, other: TrajectoryAnalyzer) -> dict[str, Any]:
"""Compare this analyzer's summary with another's.
Args:
other: Another TrajectoryAnalyzer to compare against.
Returns:
Dict showing differences:
- total_episodes_diff
- avg_reward_diff
- avg_steps_diff
- outcome_diffs
- per_role_diffs
"""
s1 = self.summary()
s2 = other.summary()
outcome_diffs: dict[str, dict[str, int]] = {}
all_outcomes = set(s1["outcomes"].keys()) | set(s2["outcomes"].keys())
for outcome in sorted(all_outcomes):
c1 = s1["outcomes"].get(outcome, 0)
c2 = s2["outcomes"].get(outcome, 0)
outcome_diffs[outcome] = {"baseline": c1, "compare": c2, "diff": c2 - c1}
per_role_diffs: dict[str, dict[str, Any]] = {}
all_roles = set(s1["per_role"].keys()) | set(s2["per_role"].keys())
for role in sorted(all_roles):
r1 = s1["per_role"].get(role, {"count": 0, "avg_reward": 0.0})
r2 = s2["per_role"].get(role, {"count": 0, "avg_reward": 0.0})
per_role_diffs[role] = {
"count_diff": r2["count"] - r1["count"],
"avg_reward_baseline": r1.get("avg_reward", 0.0),
"avg_reward_compare": r2.get("avg_reward", 0.0),
"avg_reward_diff": round(
r2.get("avg_reward", 0.0) - r1.get("avg_reward", 0.0), 4
),
}
return {
"total_episodes_diff": s2["total_episodes"] - s1["total_episodes"],
"avg_reward_baseline": s1["avg_reward"],
"avg_reward_compare": s2["avg_reward"],
"avg_reward_diff": round(s2["avg_reward"] - s1["avg_reward"], 4),
"avg_steps_baseline": s1["avg_steps"],
"avg_steps_compare": s2["avg_steps"],
"avg_steps_diff": round(s2["avg_steps"] - s1["avg_steps"], 2),
"outcome_diffs": outcome_diffs,
"per_role_diffs": per_role_diffs,
}
def report(self) -> str:
"""Generate a formatted text report.
Returns:
Multi-line string report suitable for terminal output.
"""
s = self.summary()
lines: list[str] = []
lines.append("=" * 60)
lines.append("OpenRange Trajectory Analysis Report")
lines.append("=" * 60)
lines.append("")
lines.append(f"Total episodes: {s['total_episodes']}")
lines.append(f"Total records: {s['total_records']}")
lines.append(f"Average reward: {s['avg_reward']}")
lines.append(f"Average steps: {s['avg_steps']}")
lines.append("")
# Outcomes
lines.append("Outcomes:")
for outcome, count in sorted(s["outcomes"].items()):
pct = (count / s["total_episodes"] * 100) if s["total_episodes"] > 0 else 0
lines.append(f" {outcome:<20s} {count:>5d} ({pct:.1f}%)")
lines.append("")
# Per-role stats
lines.append("Per-role statistics:")
for role, data in sorted(s["per_role"].items()):
lines.append(f" {role}:")
lines.append(f" Records: {data['count']}")
lines.append(f" Avg reward: {data['avg_reward']:.4f}")
role_outcomes = data.get("outcomes", {})
if role_outcomes:
lines.append(" Outcomes:")
for outcome, count in sorted(role_outcomes.items()):
lines.append(f" {outcome}: {count}")
lines.append("")
# Vuln class breakdown
vuln_data = self.by_vuln_class()
if vuln_data:
lines.append("Vulnerability class breakdown:")
for vc, stats in vuln_data.items():
lines.append(
f" {vc:<25s} "
f"attempts={stats['attempts']:>3d} "
f"solves={stats['solves']:>3d} "
f"rate={stats['solve_rate']:.2%}"
)
lines.append("")
lines.append("=" * 60)
return "\n".join(lines)
# ---------------------------------------------------------------------------
# CLI
# ---------------------------------------------------------------------------
def main() -> None:
parser = argparse.ArgumentParser(
description="Analyze OpenRange trajectory JSONL files",
)
parser.add_argument(
"files",
nargs="+",
help="One or more JSONL trajectory files",
)
parser.add_argument(
"--compare",
action="store_true",
help="Compare two files (requires exactly 2 file args)",
)
parser.add_argument(
"--json",
action="store_true",
dest="json_output",
help="Output summary as JSON instead of formatted report",
)
args = parser.parse_args()
if args.compare:
if len(args.files) != 2:
print("--compare requires exactly 2 files", file=sys.stderr)
sys.exit(1)
a1 = TrajectoryAnalyzer()
a1.load(args.files[0])
a2 = TrajectoryAnalyzer()
a2.load(args.files[1])
diff = a1.compare(a2)
print(json.dumps(diff, indent=2))
sys.exit(0)
analyzer = TrajectoryAnalyzer()
for f in args.files:
analyzer.load(f)
if args.json_output:
print(json.dumps(analyzer.summary(), indent=2, default=str))
else:
print(analyzer.report())
if __name__ == "__main__":
main()
|