146 lines
5.5 KiB
Python
146 lines
5.5 KiB
Python
from __future__ import annotations
|
|
|
|
import json
|
|
from dataclasses import asdict, dataclass
|
|
from pathlib import Path
|
|
from typing import Any, Protocol
|
|
|
|
from .models import BehaviorSkeleton
|
|
from .skeletons import generate_skeletons
|
|
|
|
|
|
class JsonModel(Protocol):
|
|
def request_json(
|
|
self,
|
|
*,
|
|
system: str,
|
|
user: str,
|
|
schema_name: str,
|
|
schema: dict[str, Any],
|
|
) -> dict[str, Any]: ...
|
|
|
|
def request_text(self, *, system: str, user: str) -> str: ...
|
|
|
|
|
|
@dataclass(frozen=True)
|
|
class EvaluationResult:
|
|
skeleton_id: str
|
|
family: str
|
|
complexity: int
|
|
rendered_problem: str
|
|
answer_options: tuple[str, ...]
|
|
gold_answer: str
|
|
gold_present_probability: float
|
|
solver_output: str
|
|
model_answer: str
|
|
correct: bool
|
|
|
|
|
|
RENDER_SCHEMA = {
|
|
"type": "object",
|
|
"properties": {"problem": {"type": "string"}},
|
|
"required": ["problem"],
|
|
"additionalProperties": False,
|
|
}
|
|
|
|
|
|
def render_skeleton(client: JsonModel, skeleton: BehaviorSkeleton) -> str:
|
|
prompt = json.dumps(skeleton.renderer_payload(), ensure_ascii=False, indent=2)
|
|
response = client.request_json(
|
|
system=(
|
|
"You turn behavioral observations from an unknown stochastic system into one self-contained "
|
|
"natural-language inductive reasoning problem. Invent a coherent concrete or fictional domain. "
|
|
"Give each event code and attribute a stable surface name, and preserve every case's attributes, "
|
|
"event order, trial count, observed outcome counts, query, and exact answer labels. The histories "
|
|
"are empirical observations, not explicit system laws: do not state or invent the hidden rules, "
|
|
"trigger probabilities, intermediate flags, exact query probability, or answer. Make clear that "
|
|
"all trials use one unchanged system and that the question asks which final outcome is more likely, "
|
|
"not for an exact probability. End by requiring exactly one supplied answer label."
|
|
),
|
|
user=f"Render this intermediate representation:\n{prompt}",
|
|
schema_name="rendered_problem",
|
|
schema=RENDER_SCHEMA,
|
|
)
|
|
problem = response.get("problem")
|
|
if not isinstance(problem, str) or not problem.strip():
|
|
raise RuntimeError("renderer returned an empty problem")
|
|
return problem.strip()
|
|
|
|
|
|
def parse_marked_answer(output: str, answer_options: tuple[str, ...]) -> str:
|
|
lines = [line.strip() for line in output.splitlines() if line.strip()]
|
|
if not lines or not lines[-1].startswith("####"):
|
|
raise RuntimeError("solver output must end with '#### <answer>'")
|
|
answer = lines[-1][4:].strip()
|
|
if answer not in answer_options:
|
|
raise RuntimeError(f"solver returned invalid answer label: {answer!r}")
|
|
return answer
|
|
|
|
|
|
def solve_problem(client: JsonModel, problem: str, answer_options: tuple[str, ...]) -> tuple[str, str]:
|
|
output = client.request_text(
|
|
system=(
|
|
"Infer the stable hidden mechanism from the repeated behavioral observations, then predict which "
|
|
"query outcome is more likely. Event order, attributes, stochastic frequencies, interactions, and "
|
|
"counterfactual controls may matter. Do not demand an exact probability. Give a concise explanation, then put "
|
|
"the final answer on the last non-empty line in the exact form '#### <answer>', where <answer> "
|
|
"is exactly one permitted answer label. Do not use unstated real-world assumptions."
|
|
),
|
|
user=f"{problem}\n\nPermitted answer labels: {', '.join(answer_options)}",
|
|
)
|
|
return output, parse_marked_answer(output, answer_options)
|
|
|
|
|
|
def _write_jsonl(path: Path, records: list[dict[str, Any]]) -> None:
|
|
with path.open("w", encoding="utf-8") as handle:
|
|
for record in records:
|
|
handle.write(json.dumps(record, ensure_ascii=False) + "\n")
|
|
|
|
|
|
def run_evaluation(
|
|
*,
|
|
renderer: JsonModel,
|
|
solver: JsonModel,
|
|
count: int,
|
|
length: str | int,
|
|
seed: int,
|
|
output_dir: Path,
|
|
) -> dict[str, Any]:
|
|
output_dir.mkdir(parents=True, exist_ok=True)
|
|
skeletons = generate_skeletons(count=count, length=length, seed=seed)
|
|
_write_jsonl(output_dir / "skeletons.jsonl", [item.to_dict() for item in skeletons])
|
|
|
|
results: list[EvaluationResult] = []
|
|
for skeleton in skeletons:
|
|
problem = render_skeleton(renderer, skeleton)
|
|
solver_output, answer = solve_problem(solver, problem, skeleton.answer_options)
|
|
results.append(
|
|
EvaluationResult(
|
|
skeleton_id=skeleton.id,
|
|
family=skeleton.family,
|
|
complexity=skeleton.complexity,
|
|
rendered_problem=problem,
|
|
answer_options=skeleton.answer_options,
|
|
gold_answer=skeleton.gold_answer,
|
|
gold_present_probability=skeleton.gold_present_probability,
|
|
solver_output=solver_output,
|
|
model_answer=answer,
|
|
correct=answer == skeleton.gold_answer,
|
|
)
|
|
)
|
|
|
|
_write_jsonl(output_dir / "results.jsonl", [asdict(item) for item in results])
|
|
correct = sum(item.correct for item in results)
|
|
summary = {
|
|
"count": len(results),
|
|
"correct": correct,
|
|
"accuracy": correct / len(results),
|
|
"length": length,
|
|
"complexity": skeletons[0].complexity,
|
|
"seed": seed,
|
|
}
|
|
(output_dir / "summary.json").write_text(
|
|
json.dumps(summary, ensure_ascii=False, indent=2) + "\n", encoding="utf-8"
|
|
)
|
|
return summary
|