第十一节：API、AKShare与事件数据管线

python -m pip install -U akshare pandas requests python-dotenv
python -m pip freeze > requirements-agent.txt
python -c "import akshare, pandas, requests; print(akshare.__version__)"

import akshare as ak
import pandas as pd

stocks = ak.stock_info_a_code_name()

print("对象类型：", type(stocks))
print("表格形状（行数，列数）：", stocks.shape)
print("列名：", stocks.columns.tolist())
display(stocks.head())

stock_code = "sz000001"        # 本例：深市代码，前缀和数字一起传入
start_date = "20250101"        # 起始日：YYYYMMDD
end_date = "20251231"          # 结束日：YYYYMMDD
adjust = "qfq"                 # 前复权；研究记录中必须写明

price_raw = ak.stock_zh_a_daily(
    symbol=stock_code,
    start_date=start_date,
    end_date=end_date,
    adjust=adjust,
)

print("原始表格的列名：", price_raw.columns.tolist())
print("原始表格的形状：", price_raw.shape)
display(price_raw.head(3))
display(price_raw.tail(3))

required_price_cols = ["date", "open", "high", "low", "close", "volume"]
missing_cols = [c for c in required_price_cols if c not in price_raw.columns]
if missing_cols:
    raise ValueError(f"行情接口缺少必要字段：{missing_cols}")

price = price_raw.copy()
price["date"] = pd.to_datetime(price["date"], errors="coerce")
for col in ["open", "high", "low", "close", "volume"]:
    price[col] = pd.to_numeric(price[col], errors="coerce")

print("日期缺失数：", price["date"].isna().sum())
print("收盘价缺失数：", price["close"].isna().sum())
print("重复日期数：", price["date"].duplicated().sum())
print("日期范围：", price["date"].min().date(), "至", price["date"].max().date())
print("收盘价是否全部大于 0：", (price["close"] > 0).all())

price = (
    price.dropna(subset=["date", "close"])
         .sort_values("date")
         .drop_duplicates("date")
         .reset_index(drop=True)
)

price["return_1d"] = price["close"].pct_change()
price["return_1d_pct"] = (price["return_1d"] * 100).round(2)

latest = price.iloc[-1]
previous = price.iloc[-2]
price_window = price[["date", "close", "return_1d_pct", "volume"]].tail(10).copy()

print(f"最近交易日：{latest['date'].date()}")
print(f"最新收盘价：{latest['close']:.2f}")
print(f"相对上一交易日变动：{latest['return_1d_pct']:.2f}%")
display(price_window)

news_raw = ak.stock_info_global_news()

print("新闻表格形状：", news_raw.shape)
print("新闻表格列名：", news_raw.columns.tolist())
display(news_raw.head(5))

candidate_news_cols = ["title", "publish_time", "source", "url"]
available_news_cols = [c for c in candidate_news_cols if c in news_raw.columns]

if "title" not in available_news_cols:
    raise ValueError("新闻接口没有 title 字段；请先查看列名和官方说明。")

news = news_raw[available_news_cols].copy()
if "publish_time" in news.columns:
    news["publish_time"] = pd.to_datetime(news["publish_time"], errors="coerce")

news = news.dropna(subset=["title"]).drop_duplicates(subset=["title"])
news_window = news.head(6).copy()

print("保留字段：", available_news_cols)
print("可用于摘要的新闻数：", len(news_window))
display(news_window)

price_for_prompt = price_window.copy()
price_for_prompt["date"] = price_for_prompt["date"].dt.strftime("%Y-%m-%d")
price_for_prompt = price_for_prompt.rename(columns={
    "date": "日期",
    "close": "收盘价",
    "return_1d_pct": "单日收益率(%)",
    "volume": "成交量",
})

price_text = price_for_prompt.to_string(index=False)
print(price_text)

def format_one_news(row):
    title = str(row["title"]).strip()
    time_text = "时间未提供"
    source_text = "来源未提供"

    if "publish_time" in row.index and pd.notna(row["publish_time"]):
        time_text = row["publish_time"].strftime("%Y-%m-%d %H:%M")
    if "source" in row.index and pd.notna(row["source"]):
        source_text = str(row["source"]).strip()

    return f"- [{time_text}] [{source_text}] {title}"

news_text = "\n".join(
    format_one_news(row) for _, row in news_window.iterrows()
)
print(news_text)

data_cutoff = str(price["date"].max().date())
context = f"""【数据范围】
股票代码：{stock_code}
价格区间：{price["date"].min().date()} 至 {data_cutoff}
价格口径：日线，adjust={adjust}
新闻条数：{len(news_window)}

【最近十个交易日行情】
{price_text}

【最新新闻线索】
{news_text}

【使用边界】
新闻为候选事件线索；未完成公司匹配、原文核验或因果识别。
所有文字仅用于课程研究练习，不构成投资建议。
"""
print(context)

from pathlib import Path
from datetime import datetime

Path("data").mkdir(exist_ok=True)
Path("outputs").mkdir(exist_ok=True)

run_id = datetime.now().strftime("%Y%m%d_%H%M%S")
price_window.to_csv(f"data/{run_id}_price_window.csv", index=False, encoding="utf-8-sig")
news_window.to_csv(f"data/{run_id}_news_window.csv", index=False, encoding="utf-8-sig")
Path(f"outputs/{run_id}_context.txt").write_text(context, encoding="utf-8")

print("已保存本次运行的三个文件，编号：", run_id)