File size: 9,383 Bytes
83136ac | 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 | """Evaluate scripted baselines across every curriculum tier.
Produces two artifacts (both required for the pitch):
* ``baseline.json`` β machine-readable results for every (policy, tier, seed).
* ``baseline_curve.png`` β Random vs. Heuristic vs. Oracle mean reward by tier.
Matplotlib is imported lazily so running the script without mpl still
produces the JSON.
This is the "before training" numeric evidence the rubric asks for under
criterion #3, "Showing Improvement in Rewards". The "after training"
artifact will come from :mod:`chaosops.train.grpo_train`.
"""
from __future__ import annotations
import argparse
import json
import statistics
import sys
from dataclasses import asdict, dataclass
from pathlib import Path
from typing import Callable
from chaosops.agents.policies import (
Policy,
heuristic_policy,
oracle_policy,
random_policy,
)
from chaosops.agents.runner import EpisodeResult, run_episode
from chaosops.curriculum.generator import scenarios_for_tier
from chaosops.env.environment import ChaosOpsEnvironment
from chaosops.env.models import AgentRole, DifficultyTier, FailureType
from chaosops.env.world_sim import Scenario
# ---------------------------------------------------------------------------
# Result types
# ---------------------------------------------------------------------------
@dataclass
class EpisodeStat:
policy: str
tier: str
failure_type: str
seed: int
resolved: bool
steps: int
cumulative_reward: float
wrong_fixes: int
oversight_flags: list[str]
@dataclass
class AggregateStat:
policy: str
tier: str
episodes: int
mean_reward: float
median_reward: float
resolution_rate: float
mean_mttr: float # averaged only over resolved episodes
rogue_catch_rate: float # fraction of rogue-scenarios where flag is correct
# ---------------------------------------------------------------------------
# Policy factories keyed by name
# ---------------------------------------------------------------------------
def _build_policy_for_scenario(name: str, scenario: Scenario) -> Policy:
if name == "random":
return random_policy(seed=scenario.seed)
if name == "heuristic":
return heuristic_policy(seed=scenario.seed)
if name == "oracle":
return oracle_policy(scenario.failure_type)
raise ValueError(f"unknown policy '{name}'")
PolicyFactory = Callable[[str, Scenario], Policy]
# ---------------------------------------------------------------------------
# Evaluation
# ---------------------------------------------------------------------------
def evaluate(
*,
tiers: list[DifficultyTier],
policy_names: list[str],
episodes_per_type: int,
factory: PolicyFactory = _build_policy_for_scenario,
) -> tuple[list[EpisodeStat], list[AggregateStat]]:
env = ChaosOpsEnvironment()
per_episode: list[EpisodeStat] = []
for tier in tiers:
scenarios = scenarios_for_tier(tier, episodes_per_type=episodes_per_type)
for policy_name in policy_names:
for scen in scenarios:
policy = factory(policy_name, scen)
result: EpisodeResult = run_episode(
env, scen, {r: policy for r in AgentRole}
)
per_episode.append(
EpisodeStat(
policy=policy_name,
tier=tier.value,
failure_type=scen.failure_type.value,
seed=scen.seed,
resolved=result.resolved,
steps=result.final_step,
cumulative_reward=result.cumulative_reward,
wrong_fixes=result.wrong_fixes,
oversight_flags=list(result.oversight_flags),
)
)
aggregates = _aggregate(per_episode)
return per_episode, aggregates
def _aggregate(per_episode: list[EpisodeStat]) -> list[AggregateStat]:
buckets: dict[tuple[str, str], list[EpisodeStat]] = {}
for ep in per_episode:
buckets.setdefault((ep.policy, ep.tier), []).append(ep)
out: list[AggregateStat] = []
for (policy, tier), eps in buckets.items():
rewards = [e.cumulative_reward for e in eps]
resolved = [e for e in eps if e.resolved]
mttr = (
statistics.mean(e.steps for e in resolved)
if resolved
else float("nan")
)
rogue_eps = [
e
for e in eps
if FailureType(e.failure_type).is_rogue_agent
]
rogue_catches = [
e
for e in rogue_eps
if _expected_flag(FailureType(e.failure_type)) in e.oversight_flags
]
rogue_rate = (len(rogue_catches) / len(rogue_eps)) if rogue_eps else 0.0
out.append(
AggregateStat(
policy=policy,
tier=tier,
episodes=len(eps),
mean_reward=statistics.mean(rewards),
median_reward=statistics.median(rewards),
resolution_rate=len(resolved) / len(eps),
mean_mttr=mttr,
rogue_catch_rate=rogue_rate,
)
)
return out
def _expected_flag(failure_type: FailureType) -> str:
return {
FailureType.AUTOSCALER_COST_CUT: "autoscaler",
FailureType.MISROUTED_TRAFFIC: "load_balancer",
FailureType.ROGUE_DEPLOY_BOT: "deploy_bot",
}[failure_type]
# ---------------------------------------------------------------------------
# Persistence
# ---------------------------------------------------------------------------
def save_json(
path: Path,
per_episode: list[EpisodeStat],
aggregates: list[AggregateStat],
) -> None:
path.parent.mkdir(parents=True, exist_ok=True)
payload = {
"per_episode": [asdict(e) for e in per_episode],
"aggregates": [asdict(a) for a in aggregates],
}
path.write_text(json.dumps(payload, indent=2))
def save_plot(path: Path, aggregates: list[AggregateStat]) -> bool:
"""Render the reward-vs-tier plot. Returns False if matplotlib is missing."""
try:
import matplotlib
matplotlib.use("Agg")
import matplotlib.pyplot as plt
except ImportError:
return False
tiers = [t.value for t in DifficultyTier]
policies = sorted({a.policy for a in aggregates})
color_map = {"random": "#c0392b", "heuristic": "#2980b9", "oracle": "#27ae60"}
fig, ax = plt.subplots(figsize=(8, 4.5), dpi=150)
for policy in policies:
xs, ys = [], []
for tier in tiers:
match = next(
(a for a in aggregates if a.policy == policy and a.tier == tier),
None,
)
if match is None:
continue
xs.append(tier)
ys.append(match.mean_reward)
ax.plot(xs, ys, marker="o", label=policy, color=color_map.get(policy), linewidth=2.2)
ax.axhline(0, color="#888", linewidth=0.6)
ax.set_title("ChaosOps AI β Mean Episode Reward by Difficulty Tier", fontsize=13)
ax.set_xlabel("Difficulty tier")
ax.set_ylabel("Mean cumulative reward")
ax.grid(True, linestyle=":", alpha=0.4)
ax.legend(loc="lower left")
path.parent.mkdir(parents=True, exist_ok=True)
fig.tight_layout()
fig.savefig(path)
plt.close(fig)
return True
# ---------------------------------------------------------------------------
# CLI
# ---------------------------------------------------------------------------
def _parse_args() -> argparse.Namespace:
parser = argparse.ArgumentParser()
parser.add_argument(
"--episodes-per-type",
type=int,
default=5,
help="episodes per (tier, failure type)",
)
parser.add_argument(
"--out-dir",
type=Path,
default=Path("artifacts/baseline"),
)
parser.add_argument(
"--policies",
nargs="+",
default=["random", "heuristic", "oracle"],
choices=["random", "heuristic", "oracle"],
)
return parser.parse_args()
def main() -> None:
args = _parse_args()
tiers = list(DifficultyTier)
per_episode, aggregates = evaluate(
tiers=tiers,
policy_names=args.policies,
episodes_per_type=args.episodes_per_type,
)
save_json(args.out_dir / "baseline.json", per_episode, aggregates)
rendered = save_plot(args.out_dir / "baseline_curve.png", aggregates)
print(f"wrote {args.out_dir / 'baseline.json'}")
if rendered:
print(f"wrote {args.out_dir / 'baseline_curve.png'}")
else:
print("matplotlib not installed β skipping PNG render", file=sys.stderr)
print()
print(f"{'policy':<10} {'tier':<8} {'eps':>4} {'mean_R':>9} {'med_R':>9} "
f"{'res%':>6} {'mttr':>6} {'rogue_catch':>12}")
print("-" * 80)
for a in sorted(aggregates, key=lambda x: (x.policy, x.tier)):
mttr = f"{a.mean_mttr:.1f}" if a.mean_mttr == a.mean_mttr else "β"
print(
f"{a.policy:<10} {a.tier:<8} {a.episodes:>4} "
f"{a.mean_reward:>+9.1f} {a.median_reward:>+9.1f} "
f"{a.resolution_rate:>5.0%} {mttr:>6} {a.rogue_catch_rate:>11.0%}"
)
if __name__ == "__main__":
main()
|