File size: 3,169 Bytes
2b9a95b | 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 | """
Standalone CausalGame evaluator.
Loads a scenario's SCM + config directly from the CausalGame repo (no HTTP server)
and evaluates a drone design on a fleet, returning the survival rate. This mirrors
exactly what api/modules/agent/action_space.AgentActionSpace.execute() does for the
Stage-2 final evaluation, so numbers match the live backend.
"""
import os
import sys
import random
from pathlib import Path
CG_ROOT = Path(os.environ.get("CG_ROOT", Path(__file__).resolve().parents[1] / "CausalGame"))
sys.path.insert(0, str(CG_ROOT))
from api.app import load_experiment_config # noqa: E402
from api.modules.environment.scm_registry import get_scm_for_experiment # noqa: E402
from api.modules.agent.action_space import AgentActionSpace, SubmitAction # noqa: E402
DEFAULT_DESIGN = {
"engine_def": 20, "cockpit_def": 20, "wing_def": 15,
"body_def": 15, "antenna_def": 10, "camera_def": 5, "gun_def": 5,
}
def make_action_space(experiment: str):
cfg = load_experiment_config(experiment)
try:
scm = get_scm_for_experiment(experiment, cfg)
except ValueError:
# Some folder variants (e.g. *_categorical_no_selection_bias) are not
# separately registered in the SCM registry, but the Stage-2 ground-truth
# SCM is identical to the family base. Fall back to the base SCM while
# keeping the scenario's own game.json config (thresholds, params).
base = experiment
scm = None
while "_" in base:
base = base.rsplit("_", 1)[0]
try:
scm = get_scm_for_experiment(base, cfg)
break
except ValueError:
continue
if scm is None:
raise
return AgentActionSpace(scm, cfg), cfg
def evaluate(experiment: str, design: dict, equipment: dict = None, seed: int = None):
"""Return dict(survival_rate, survived, fleet_size, victory, threshold)."""
if seed is not None:
random.seed(seed)
asp, cfg = make_action_space(experiment)
res = asp.execute(SubmitAction(design=design, equipment=equipment or {}))
if not getattr(res, "success", False):
return {"error": getattr(res, "error", "unknown")}
return {
"survival_rate": res.survival_rate,
"survived": res.survived,
"fleet_size": res.fleet_size,
"victory": res.victory,
"threshold": res.victory_threshold,
}
def evaluate_mean(experiment: str, design: dict, equipment: dict = None, seeds=range(3)):
rates = []
for s in seeds:
r = evaluate(experiment, design, equipment, seed=s)
if "error" in r:
return r
rates.append(r["survival_rate"])
import statistics
return {
"survival_mean": statistics.mean(rates),
"survival_std": statistics.pstdev(rates) if len(rates) > 1 else 0.0,
"rates": rates,
"threshold": r["threshold"],
"victory_mean": statistics.mean(rates) >= r["threshold"],
}
if __name__ == "__main__":
exp = sys.argv[1] if len(sys.argv) > 1 else "antenna_trap"
print("Scenario:", exp)
print("Default design:", evaluate_mean(exp, DEFAULT_DESIGN))
|