mirror of
https://github.com/shy3130/tick-stock-panel.git
synced 2026-11-02 10:26:13 +00:00
shy3130
918c7ade7e
- 策略: 引擎只加载 data/strategies/{custom,ai,composite} 与研究模板目录,
原 25 个内置策略一次性迁移为用户自定义文件(ID 不变, 引用零迁移);
研究模板移至 app/strategy/research, 打包 spec 同步; 新装环境策略库为空
- 信号: 管线不再硬编码任何信号表达式; 涨跌停判定价成为运行时列
limit_up_price/limit_down_price(生效价∓0.005 容差), 20 个信号转为
data/user_data/custom_signals 定义(显式 column 沿用 signal_* 原列名);
依赖解析/注入顺序/当日路径 prev-join 全部定义驱动; 语义统一为历史路径
口径(顺带修正当日路径 BOLL 比较符与翘板/炸板空值两处既有偏差)
- 前端: 信号库单列表(全部可编辑), SignalPicker/监控选项/信号标签全走
定义接口; 回测页新增免责声明(模拟性质/数据源准确性不作保证/不构成
投资建议)
- 测试: 25 个策略夹具与 20 个信号种子入库 tests/fixtures, 受影响测试
全部自包含化(不依赖运行目录用户数据); 77 万行 MA 死叉新旧表达式对拍
逐行一致; 回测实跑 45 笔卖点信号出场验证矩阵链路
- 同批包含工作区既有进度: 策略结果 ext 列同日投影修复(含测试)、数据源插件、fuyao/auth/api_gateway/data_integrity 调整、README 信号库
描述与 strategy.md 导入路径补充
验证: 后端 2592 passed(空数据目录 CI 模拟 2588 passed, 4 例为模拟环境
伪影复验排除); 前端 vitest 125 passed + tsc 0 错 + vite build 通过
337 lines
12 KiB
Python
337 lines
12 KiB
Python
# -*- coding: utf-8 -*-
|
|
"""AI 策略迭代闭环的定向测试 (stub 生成器与工具循环, 不依赖真实 AI key 与数据)。
|
|
|
|
覆盖三层:
|
|
1. _rewrite_meta_id 纯函数 (三种 META 声明形式 + 缺失 id 插入)
|
|
2. tool_catalog.build_tool_schemas + execute_tool 的 list_* 分发与未知工具
|
|
3. AIStrategyIterator.iterate 完整状态流转 (收敛 / 校验失败 / 末轮改进补 final 回测)
|
|
"""
|
|
from __future__ import annotations
|
|
|
|
import ast
|
|
import asyncio
|
|
import json
|
|
import sys
|
|
import tempfile
|
|
import types
|
|
from pathlib import Path
|
|
|
|
from app.services import tool_catalog
|
|
from app.strategy import ai_iterator as it
|
|
from app.strategy.ai_generator import AIStrategyGenerator as RealGen
|
|
from app.strategy.ai_generator import find_meta_assignment
|
|
from app.strategy.ai_iterator import AIStrategyIterator, _rewrite_meta_id
|
|
|
|
V1_CODE = '''"""v1 策略"""
|
|
import polars as pl
|
|
|
|
META = {
|
|
"id": "ai_placeholder",
|
|
"name": "v1",
|
|
"description": "v1",
|
|
"tags": ["ai"],
|
|
"asset_types": ["stock"],
|
|
"timeframes": ["1d"],
|
|
"params": [],
|
|
"scoring": {"change_pct": 1.0},
|
|
"order_by": "score",
|
|
"descending": True,
|
|
"limit": 100,
|
|
}
|
|
|
|
EXECUTION_BACKEND = "polars_expr"
|
|
|
|
def filter(df: pl.DataFrame, params: dict) -> pl.Expr:
|
|
return pl.col("close") > pl.col("ma5")
|
|
'''
|
|
|
|
# v2 与 v1 不同: 增加成交量条件
|
|
V2_CODE = V1_CODE.replace(
|
|
'return pl.col("close") > pl.col("ma5")',
|
|
'return (pl.col("close") > pl.col("ma5")) & (pl.col("volume") > pl.col("vol_ma5") * 1.2)',
|
|
).replace('"name": "v1"', '"name": "v2"')
|
|
|
|
META = {
|
|
"name": "v1",
|
|
"description": "v1",
|
|
"tags": ["ai"],
|
|
"asset_types": ["stock"],
|
|
"timeframes": ["1d"],
|
|
"params": [],
|
|
"scoring": {"change_pct": 1.0},
|
|
"order_by": "score",
|
|
"descending": True,
|
|
"limit": 100,
|
|
}
|
|
|
|
|
|
def _meta_id_of(code: str) -> str:
|
|
"""从代码里提取 META 的 id (走真实 find_meta_assignment + literal_eval)。"""
|
|
found = find_meta_assignment(code)
|
|
assert found is not None, "找不到 META"
|
|
_, value = found
|
|
return ast.literal_eval(value)["id"]
|
|
|
|
|
|
# ── Part 1: _rewrite_meta_id ──────────────────────────────
|
|
|
|
def test_rewrite_multiline_assign():
|
|
out = _rewrite_meta_id(V1_CODE, "ai_abcd1234", META)
|
|
ast.parse(out) # 必须仍是合法 Python
|
|
assert _meta_id_of(out) == "ai_abcd1234"
|
|
assert "META = {" in out
|
|
assert "STRATEGY_META" not in out
|
|
|
|
|
|
def test_rewrite_normalize_variable_name():
|
|
code = V1_CODE.replace("META = {", "STRATEGY_META = {")
|
|
out = _rewrite_meta_id(code, "ai_x", META)
|
|
ast.parse(out)
|
|
assert _meta_id_of(out) == "ai_x"
|
|
assert "STRATEGY_META" not in out
|
|
assert "META = {" in out
|
|
|
|
|
|
def test_rewrite_annassign():
|
|
code = V1_CODE.replace("META = {", "META: dict = {")
|
|
out = _rewrite_meta_id(code, "ai_y", META)
|
|
ast.parse(out)
|
|
assert _meta_id_of(out) == "ai_y"
|
|
|
|
|
|
def test_rewrite_insert_missing_id():
|
|
code = V1_CODE.replace(' "id": "ai_placeholder",\n', "")
|
|
out = _rewrite_meta_id(code, "ai_z", META)
|
|
ast.parse(out)
|
|
assert _meta_id_of(out) == "ai_z"
|
|
|
|
|
|
# ── Part 2: tool_catalog ──────────────────────────────────
|
|
|
|
def _stub_module(monkeypatch, fullname: str, **attrs) -> types.ModuleType:
|
|
"""往 sys.modules 塞 stub 模块 (monkeypatch 在测试后自动恢复)。"""
|
|
mod = types.ModuleType(fullname)
|
|
for k, v in attrs.items():
|
|
setattr(mod, k, v)
|
|
monkeypatch.setitem(sys.modules, fullname, mod)
|
|
parts = fullname.split(".")
|
|
for i in range(1, len(parts)):
|
|
parent = ".".join(parts[:i])
|
|
if parent not in sys.modules:
|
|
pkg = types.ModuleType(parent)
|
|
pkg.__path__ = []
|
|
monkeypatch.setitem(sys.modules, parent, pkg)
|
|
return mod
|
|
|
|
|
|
class _FakeFactorSpec:
|
|
def __init__(self, fid):
|
|
self.id = fid
|
|
self.label = "因子"
|
|
self.group = "g"
|
|
self.formula_text = "f"
|
|
self.kind = "k"
|
|
self.dependencies = {"close"}
|
|
self.asset_types = {"stock"}
|
|
self.stability = "stable"
|
|
self.pit = False
|
|
|
|
|
|
def test_build_tool_schemas():
|
|
schemas = tool_catalog.build_tool_schemas()
|
|
names = [s["function"]["name"] for s in schemas]
|
|
assert names == [
|
|
"list_factors", "list_strategies", "list_data_capabilities", "run_backtest",
|
|
]
|
|
for s in schemas:
|
|
assert s["type"] == "function"
|
|
assert s["function"]["parameters"]["type"] == "object"
|
|
|
|
|
|
def test_execute_list_factors(monkeypatch):
|
|
_stub_module(
|
|
monkeypatch,
|
|
"app.factors.registry",
|
|
all_factors=lambda asset_type=None, stable_only=False: [
|
|
_FakeFactorSpec("f1"), _FakeFactorSpec("f2"),
|
|
],
|
|
)
|
|
result = asyncio.run(tool_catalog.execute_tool("list_factors", {}))
|
|
assert result["ok"] is True
|
|
assert [f["id"] for f in result["result"]["factors"]] == ["f1", "f2"]
|
|
|
|
|
|
def test_execute_list_data_capabilities(monkeypatch):
|
|
_stub_module(
|
|
monkeypatch,
|
|
"app.data_providers.capabilities",
|
|
CAPABILITY_REGISTRY=[
|
|
{"id": "daily", "label": "日K", "desc": "d", "tf_tier": 0},
|
|
],
|
|
)
|
|
result = asyncio.run(tool_catalog.execute_tool("list_data_capabilities", {}))
|
|
assert result["ok"] is True
|
|
assert result["result"]["capabilities"][0]["id"] == "daily"
|
|
|
|
|
|
class _FakeEngine:
|
|
def __init__(self):
|
|
self.reloads = 0
|
|
|
|
def has(self, sid):
|
|
return False
|
|
|
|
def reload(self):
|
|
self.reloads += 1
|
|
|
|
def list_strategies(self, include_research=False):
|
|
return [{"id": "builtin_1", "name": "内置", "description": "", "tags": [],
|
|
"asset_types": ["stock"], "timeframes": ["1d"], "execution_backend": "polars_expr",
|
|
"source": "custom", "params": []}]
|
|
|
|
|
|
def test_execute_list_strategies():
|
|
engine = _FakeEngine()
|
|
result = asyncio.run(tool_catalog.execute_tool("list_strategies", {}, engine=engine))
|
|
assert result["ok"] is True
|
|
assert result["result"]["strategies"][0]["id"] == "builtin_1"
|
|
|
|
|
|
def test_execute_unknown_tool():
|
|
result = asyncio.run(tool_catalog.execute_tool("no_such_tool", {}))
|
|
assert result["ok"] is False
|
|
assert "未知工具" in result["error"]
|
|
|
|
|
|
# ── Part 3: AIStrategyIterator.iterate 状态流转 ────────────
|
|
|
|
class FakeGenerator:
|
|
"""替换 AIStrategyGenerator: generate 返回 v1, validate_code 对改进版返回 valid。"""
|
|
|
|
def needs_structural_repair(self, result):
|
|
return False
|
|
|
|
def _get_guide(self):
|
|
return "guide"
|
|
|
|
async def generate(self, prompt):
|
|
return {"code": V1_CODE, "meta": dict(META), "valid": True, "error": None}
|
|
|
|
def validate_code(self, code):
|
|
stripped = RealGen._extract_code_block(code)
|
|
return {"code": stripped, "meta": dict(META), "valid": True, "error": None}
|
|
|
|
async def repair_code(self, code, error):
|
|
raise AssertionError("不应进入 repair_code 分支")
|
|
|
|
|
|
def _fake_tool_loop(call_log: list):
|
|
"""按调用次数返回对话: 第 1 轮 run_backtest→改进, 第 2 轮原样输出(收敛)。"""
|
|
async def fake(messages, tools, *, execute_tool, max_rounds, temperature, max_tokens):
|
|
call_log.append(len(call_log) + 1)
|
|
if len(call_log) == 1:
|
|
return [
|
|
{"role": "assistant", "content": None, "tool_calls": [{
|
|
"id": "call_1", "type": "function",
|
|
"function": {"name": "run_backtest",
|
|
"arguments": json.dumps({"strategy_id": "draft"})},
|
|
}]},
|
|
{"role": "tool", "tool_call_id": "call_1",
|
|
"content": json.dumps({"ok": True, "result": {"stats": {
|
|
"sharpe": 1.5, "max_drawdown": -0.08, "win_rate": 0.55, "total_return": 0.12,
|
|
}}}, ensure_ascii=False)},
|
|
{"role": "assistant", "content": "改进夏普\n```python\n" + V2_CODE + "\n```"},
|
|
]
|
|
return [
|
|
{"role": "assistant", "content": "已无改进空间\n```python\n" + V2_CODE + "\n```"},
|
|
]
|
|
|
|
return fake
|
|
|
|
|
|
def test_iterate_full_flow(monkeypatch):
|
|
call_log: list = []
|
|
monkeypatch.setattr(it, "AIStrategyGenerator", FakeGenerator)
|
|
monkeypatch.setattr(it, "generate_ai_text_with_tools", _fake_tool_loop(call_log))
|
|
|
|
engine = _FakeEngine()
|
|
with tempfile.TemporaryDirectory() as tmp:
|
|
result = asyncio.run(
|
|
AIStrategyIterator(max_rounds=2).iterate(
|
|
"做一个均线多头策略", engine=engine, data_dir=tmp,
|
|
)
|
|
)
|
|
draft_id = result["draft_strategy_id"]
|
|
# 在 with 块内读落盘文件 (退出后 TemporaryDirectory 会清理)
|
|
draft_path = Path(tmp) / "strategies" / "ai" / f"{draft_id}.py"
|
|
assert draft_path.exists()
|
|
on_disk = draft_path.read_text(encoding="utf-8")
|
|
|
|
assert draft_id.startswith("ai_"), draft_id
|
|
assert len(draft_id) == len("ai_") + 8
|
|
|
|
# 2 轮: 第 1 轮改进, 第 2 轮收敛 (收敛路径 final 已被回测, 不补末行)
|
|
assert len(result["rounds"]) == 2, result["rounds"]
|
|
r1, r2 = result["rounds"]
|
|
assert r1["stats"] is not None and r1["stats"]["sharpe"] == 1.5
|
|
assert r1["change_summary"] == "改进夏普"
|
|
assert r2["stats"] is None
|
|
assert "收敛" in r2["change_summary"]
|
|
|
|
assert result["final_code"].strip() == V2_CODE.strip()
|
|
assert result["final_meta"]["id"] == draft_id
|
|
|
|
# 引擎 reload: v1 保存 1 次 + 第 1 轮改进保存 1 次 = 2 次
|
|
assert engine.reloads == 2, engine.reloads
|
|
|
|
# 草稿文件 META.id 与文件名对齐, 且落盘的是改进后的 v2, research_only 已注入
|
|
assert _meta_id_of(on_disk) == draft_id
|
|
assert "vol_ma5" in on_disk
|
|
assert ast.literal_eval(find_meta_assignment(on_disk)[1]).get("research_only") is True
|
|
|
|
|
|
def test_iterate_validation_failure_breaks(monkeypatch):
|
|
"""改进版校验失败 -> 记录 error 并 break, 不落盘坏代码。"""
|
|
call_log: list = []
|
|
|
|
class BadGen(FakeGenerator):
|
|
def validate_code(self, code):
|
|
return {"code": code, "meta": {}, "valid": False, "error": "找不到策略入口函数 filter()"}
|
|
|
|
monkeypatch.setattr(it, "AIStrategyGenerator", BadGen)
|
|
monkeypatch.setattr(it, "generate_ai_text_with_tools", _fake_tool_loop(call_log))
|
|
|
|
engine = _FakeEngine()
|
|
with tempfile.TemporaryDirectory() as tmp:
|
|
result = asyncio.run(
|
|
AIStrategyIterator(max_rounds=2).iterate("p", engine=engine, data_dir=tmp)
|
|
)
|
|
assert len(result["rounds"]) == 1
|
|
assert "校验失败" in result["rounds"][0]["change_summary"]
|
|
assert engine.reloads == 1 # 只有 v1 落盘
|
|
|
|
|
|
def test_iterate_final_backtest_appended(monkeypatch):
|
|
"""末轮是改进版 (耗尽 max_rounds) -> final 从未被回测, 补一次末行「最终版回测」。"""
|
|
call_log: list = []
|
|
monkeypatch.setattr(it, "AIStrategyGenerator", FakeGenerator)
|
|
monkeypatch.setattr(it, "generate_ai_text_with_tools", _fake_tool_loop(call_log))
|
|
|
|
async def fake_final_backtest(self, data_dir, draft_id):
|
|
return {"sharpe": 2.0}
|
|
|
|
monkeypatch.setattr(AIStrategyIterator, "_backtest_final", fake_final_backtest)
|
|
|
|
engine = _FakeEngine()
|
|
with tempfile.TemporaryDirectory() as tmp:
|
|
result = asyncio.run(
|
|
AIStrategyIterator(max_rounds=1).iterate("p", engine=engine, data_dir=tmp)
|
|
)
|
|
|
|
# 1 轮改进 + 1 行最终版回测
|
|
assert len(result["rounds"]) == 2, result["rounds"]
|
|
r1, r2 = result["rounds"]
|
|
assert r1["change_summary"] == "改进夏普"
|
|
assert r2["change_summary"] == "最终版回测"
|
|
assert r2["stats"] == {"sharpe": 2.0}
|