Files
AgentLab/backend/tests/fixtures/dual_check_shop.py
T

162 lines
8.2 KiB
Python
Raw Blame History

This file contains ambiguous Unicode characters
This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.
"""双盲复算测试用的虚构零售库和盲算模型的剧本。表名、数字、地名都是编的。
订单表埋了两个陷阱:退款行(status = 'refunded'),以及下单日和支付日不在同一周的单。2026-09-01 至 2026-09-07 这一周:
- 按支付日、排除退款(口径说明要的):1、2、5 号单,330.0
- 按下单日、排除退款:2、4、5 号单,310.0
- 按支付日、不排除退款:1、2、3、5 号单,380.0
- 按下单日、不排除退款:2、3、4、5 号单,360.0
- 付款客户数(按支付日、排除退款):按行计数 3,去重计数 2
另有渠道(channel)和测试单标记(is_test,都是 0)两列:写数据目录的测试拿它们凑「指标自己的筛选范围」和标记列。
下单日、支付日两列声明为 DATE(SQLite 照样存成文字,按字符串比的先后和按日期比的一样):双盲复算只在两边的时间列都
声明为带日期的类型时才把选择记成业务日期。build(date_type="TEXT") 建一份声明成文本的、build(date_type="") 建一份
不写类型的,看不记的那几种。
"""
from __future__ import annotations
import json
import re
import sqlite3
from pathlib import Path
from langchain_core.messages import AIMessage, ToolMessage
from app.data import introspect
from app.db.base import SessionLocal
from app.db.models import DataSource
from tests.fixtures.names import letters
ROWS = [
(1, "华东", 101, 100.0, "paid", "2026-08-31", "2026-09-01", "online", 0),
(2, "华东", 102, 200.0, "paid", "2026-09-02", "2026-09-02", "offline", 0),
(3, "华南", 101, 50.0, "refunded", "2026-09-03", "2026-09-03", "online", 0),
(4, "华南", 103, 80.0, "paid", "2026-09-07", "2026-09-08", "online", 0),
(5, "华北", 102, 30.0, "paid", "2026-09-05", "2026-09-05", "offline", 0),
]
WEEK = "BETWEEN '2026-09-01' AND '2026-09-07'"
#: 口径说明要的写法:按支付日、排除退款
RIGHT = f"SELECT SUM(amount) AS gmv FROM orders WHERE status <> 'refunded' AND paid_date {WEEK}"
#: 只错在时间列:按下单日
BY_ORDER_DATE = f"SELECT SUM(amount) AS gmv FROM orders WHERE status <> 'refunded' AND order_date {WEEK}"
#: 只错在没排除退款
WITH_REFUNDS = f"SELECT SUM(amount) AS gmv FROM orders WHERE paid_date {WEEK}"
#: 两处都错
BOTH_WRONG = f"SELECT SUM(amount) AS gmv FROM orders WHERE order_date {WEEK}"
#: 付款客户数:按行计数 / 去重计数
BUYERS = f"SELECT COUNT(customer_id) AS buyers FROM orders WHERE status <> 'refunded' AND paid_date {WEEK}"
BUYERS_DISTINCT = (f"SELECT COUNT(DISTINCT customer_id) AS buyers FROM orders "
f"WHERE status <> 'refunded' AND paid_date {WEEK}")
#: 按地区分组:华东一行是 300.0(1、2 号单)
BY_REGION = (f"SELECT region, SUM(amount) AS gmv FROM orders WHERE status <> 'refunded' AND paid_date {WEEK} "
"GROUP BY region ORDER BY region")
DEFINITION = "已支付订单的金额合计:按支付日期(paid_date)归属到当周,不含退款"
def build(path: Path, *, date_type: str = "DATE") -> Path:
db = sqlite3.connect(path)
db.executescript(
"CREATE TABLE orders (id INTEGER PRIMARY KEY, region TEXT, customer_id INTEGER, amount REAL, status TEXT, "
f"order_date {date_type}, paid_date {date_type}, channel TEXT, is_test INTEGER);")
db.executemany("INSERT INTO orders VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?)", ROWS)
db.commit()
db.close()
return path
async def add_source(path: Path, *, options: dict | None = None) -> tuple[str, str]:
"""登记成只读的手工源,探查好表结构。返回 (数据源名, id)。"""
name = f"dc_shop_{letters(8)}"
async with SessionLocal() as session:
row = DataSource(name=name, kind="sqlite", database=str(path), readonly=True, enabled=True,
options=dict(options or {}))
session.add(row)
await session.flush()
row.schema_cache = await introspect.introspect(row)
await session.commit()
return name, row.id
def _text(messages) -> str:
return "\n".join(m.content if isinstance(m.content, str) else json.dumps(m.content, ensure_ascii=False)
for m in messages)
class Blind:
"""盲算模型的剧本,也管图里别的模型调用。
plans:{任务里出现的指标名: (SQL, 要交的列) 或 [SQL, …]、列}。键写成「指标名|输入2」时只认计算式里那一项的盲算
(一个指标有几个输入来自查询时用)。盲算循环里依次查这几条 SQL,收尾时交最后一条查询的第 0 行那一列(row 可改)。
没对上指标名的盲算交 null。不是盲算的调用(报告撰写)返回 writer 的正文并记下收到的消息。
calls 记下每次盲算调用(step / final),tasks 记下盲算收到的任务原文(几个盲算同时进行时也是每个任务一条)。
on_query:盲算第一次查库之前调用一次(比如往库里插一行,造出「两次取数之间数据变了」)。
rationale:盲算交的理由,{ref} 换成交的那一格的引用(缺省「按口径说明统计,结果是 {ref}。」)。
"""
def __init__(self, monkeypatch, source: str, plans: dict, *, writer: str = "本周销售额见口径卡。", row: int = 0,
on_query=None, rationale: str = "按口径说明统计,结果是 {ref}。"):
from app.providers import mock_model
self.source = source
self.plans = plans
self.writer = writer
self.row = row
self.on_query = on_query
self.rationale = rationale
self.calls: list[str] = []
self.tasks: list[str] = []
self.writer_prompts: list[str] = []
blind = self
def decide(model, messages):
text = _text(messages)
plan = blind.plan_of(text)
if model.response_format:
blind.calls.append("final")
done = len(re.findall(r"【证据 Q\d+】", text))
if plan is None or not done:
return AIMessage(content=json.dumps({"value": None, "rationale": "没有查到。"}, ensure_ascii=False))
_, column = plan
alias = f"Q{len(blind.sqls(plan))}"
body = {"value": {"alias": alias, "row": blind.row, "column": column},
"rationale": blind.rationale.format(ref=f"[[v:{alias}.r{blind.row}.{column}]]")}
return AIMessage(content=json.dumps(body, ensure_ascii=False))
if model.tools:
blind.calls.append("step")
# 盲算的任务是第一条用户消息;同一次盲算的几轮收到的是同一条,只记一次
task_text = next((str(m.content) for m in messages if m.type == "human"), "")
if task_text not in blind.tasks:
blind.tasks.append(task_text)
done = sum(isinstance(m, ToolMessage) for m in messages)
sqls = blind.sqls(plan) if plan is not None else []
if done < len(sqls):
if blind.on_query is not None:
hook, blind.on_query = blind.on_query, None
hook()
call = {"name": f"db_query__{blind.source}", "args": {"sql": sqls[done]},
"id": f"blind-{letters(6)}"}
return AIMessage(content="查一下。", tool_calls=[call])
return AIMessage(content="查完了。")
blind.writer_prompts.append(text)
return AIMessage(content=blind.writer)
monkeypatch.setattr(mock_model.MockChatModel, "_decide", decide)
def plan_of(self, text: str):
for key, plan in self.plans.items():
name, _, label = key.partition("|")
if f"指标:{name}" in text and (not label or f"计算式里的 {label}" in text):
return plan
return None
@staticmethod
def sqls(plan) -> list[str]:
sql = plan[0]
return list(sql) if isinstance(sql, (list, tuple)) else [sql]
@property
def runs(self) -> int:
"""盲算 agent 一共执行了几次(每次收尾交一次结果)。"""
return self.calls.count("final")