FinDataPilot / app /agent /presentation.py
Fin-DataPilot Deploy Bot
ci: e6c9696
6533267
Raw History Blame Contribute Delete
5.18 kB
"""Build bounded, source-backed answer visuals without asking an LLM for numbers."""
from __future__ import annotations
import math
import re
from collections.abc import Mapping, Sequence
from typing import Any, Literal
from pydantic import BaseModel, Field
class ChartPoint(BaseModel):
label: str
value: float
class AnswerChart(BaseModel):
id: str
type: Literal["bar"] = "bar"
title: str
unit: str = ""
points: list[ChartPoint]
source: str
trace_id: str
as_of: str | None = None
total_rows: int
class AnswerPresentation(BaseModel):
version: Literal[1] = 1
status: Literal["ready", "partial", "empty"]
successful_tools: int
total_tools: int
warnings: list[str] = Field(default_factory=list)
charts: list[AnswerChart] = Field(default_factory=list)
_IDENTIFIER = re.compile(r"代码|编码|序号|排名|日期|时间|code|(^|_)id$|rank|date|time", re.I)
_LABEL = re.compile(r"简称|名称|名字|日期|name|label|date", re.I)
_NUMBER = re.compile(r"^([+-]?(?:\d+(?:\.\d*)?|\.\d+))\s*(%|亿|万|千|元|股|倍)?$")
_METRICS = ("市盈率", "市净率", "股息率", "营业收入", "营收", "净利润", "市值", "涨跌幅", "成交额", "换手率", "净流入", "毛利率", "收益率")
def _number(value: Any) -> tuple[float, str] | None:
if isinstance(value, bool) or value is None:
return None
if isinstance(value, int | float):
number = float(value)
return (number, "") if math.isfinite(number) else None
if not isinstance(value, str):
return None
match = _NUMBER.fullmatch(value.strip().replace(",", "").replace(",", "").replace("%", "%"))
if not match:
return None
number = float(match[1])
return (number, match[2] or "") if math.isfinite(number) else None
def _charts(call: Mapping[str, Any], limit: int) -> list[AnswerChart]:
# Only structured financial records are comparable; news text is never
# mined for numbers, and identifiers are never treated as measurements.
if call.get("name") != "financial-query" or not call.get("ok"):
return []
result = call.get("result")
if not isinstance(result, dict):
return []
data = result.get("data")
raw_rows = data.get("datas", data.get("rows")) if isinstance(data, dict) else data
if not isinstance(raw_rows, list):
return []
rows = [row for row in raw_rows if isinstance(row, dict)]
if len(rows) < 2:
return []
labels = [key for key in rows[0] if _LABEL.search(key)]
if not labels:
return []
label_key = labels[0]
selected = rows[:12]
# Ambiguous labels make the visual misleading. Keep the evidence table.
names = [str(row.get(label_key) or "").strip() for row in selected]
if not all(names) or len(set(names)) != len(names):
return []
charts: list[AnswerChart] = []
query = str((call.get("args") or {}).get("query") or "")
preferred = [key for key in rows[0] if any(metric in query and metric in key for metric in _METRICS)]
# Vendors often prepend latest price/change columns to every response.
# When the user asked for a particular metric, visualise that evidence first.
for key in preferred or list(rows[0]):
if key == label_key or _IDENTIFIER.search(key):
continue
values = [_number(row.get(key)) for row in selected]
if any(value is None for value in values):
continue
valid = [value for value in values if value is not None]
units = {value[1] for value in valid}
if len(units) != 1:
continue
unit = valid[0][1]
charts.append(AnswerChart(
id=f"{call.get('trace_id', 'chart')}:{key}",
title=key,
unit=unit,
points=[ChartPoint(label=name, value=value[0]) for name, value in zip(names, valid, strict=True)],
source=str(result.get("source") or "问财金融数据"),
trace_id=str(call.get("trace_id") or ""),
as_of=result.get("as_of"),
total_rows=len(rows),
))
if len(charts) >= limit:
break
return charts
def build_presentation(
calls: Sequence[Mapping[str, Any]], end_note: str = "",
) -> dict[str, Any]:
"""Use only successful tool payloads; cap the output at three charts."""
successful = sum(call.get("ok") is True for call in calls)
warnings = [
f"{call.get('name', '数据查询')} 未成功,相关结论缺少该项证据。"
for call in calls if not call.get("ok")
]
if end_note:
warnings.append(end_note)
charts: list[AnswerChart] = []
for call in calls:
if len(charts) >= 3:
break
charts.extend(_charts(call, 3 - len(charts)))
status: Literal["ready", "partial", "empty"] = "ready"
if not successful:
status = "empty"
elif warnings:
status = "partial"
return AnswerPresentation(
status=status,
successful_tools=successful,
total_tools=len(calls),
warnings=list(dict.fromkeys(warnings))[:8],
charts=charts,
).model_dump()