#!/usr/bin/env python3
# -*- coding: utf-8 -*-
"""
Towin_robot.py
多股票量化策略总报告生成器（修复版）
功能：解决PDF黑色块问题+自动识别表头+内存流传图+同目录字体加载+PDF防乱码
"""
import tushare as ts # 交易日历
from sklearn.linear_model import LogisticRegression
from sklearn.metrics import roc_auc_score
import joblib, pickle, datetime as dt
import sys
import os
import re
import time
import warnings
from io import BytesIO
from datetime import datetime
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
# ReportLab相关（中文字体支持核心）
from reportlab.pdfgen import canvas
from reportlab.lib.pagesizes import A4
from reportlab.lib import colors
from reportlab.lib.styles import getSampleStyleSheet, ParagraphStyle
from reportlab.platypus import SimpleDocTemplate, Table, TableStyle, Paragraph, Spacer, Image
from reportlab.lib.utils import ImageReader
from reportlab.pdfbase import pdfmetrics
from reportlab.pdfbase.ttfonts import TTFont
# PDF合并
from PyPDF2 import PdfMerger
# 屏蔽警告
warnings.filterwarnings("ignore")
# -------------------------------------------------
# 核心修复：中文字体初始化（适配脚本同目录下的simhei.ttf）
# -------------------------------------------------
def init_chinese_font():
"""初始化中文字体（优先加载脚本同目录下的simhei.ttf）"""
font_name = "SimHei"
script_dir = os.path.dirname(os.path.abspath(__file__))
font_path = os.path.join(script_dir, "simhei.ttf")
if not os.path.exists(font_path):
print(f"❌ 未找到字体文件：{font_path}")
print(f" 请将simhei.ttf字体文件放在与脚本同一目录下（当前目录：{script_dir}）")
sys.exit(1)
try:
pdfmetrics.registerFont(TTFont(font_name, font_path))
print(f"✅ 成功注册中文字体：{font_name}（路径：{font_path}）")
return font_name
except Exception as e:
print(f"❌ 字体注册失败：{str(e)}")
sys.exit(1)
# -------------------------------------------------
# Matplotlib图表中文字体设置（防图表乱码）
# -------------------------------------------------
def init_matplotlib_font(font_name):
"""初始化Matplotlib字体（确保图表中文正常显示）"""
plt.rcParams.update({
'font.sans-serif': [font_name, 'DejaVu Sans'],
'axes.unicode_minus': False,
'figure.dpi': 100,
'savefig.facecolor': 'white'
})
print("✅ Matplotlib图表字体初始化完成")
# -------------------------------------------------
# 工具函数：证券代码提取+Excel文件获取
# -------------------------------------------------
def extract_stock_code(filename: str) -> str:
"""从文件名提取6位证券代码，失败则生成唯一哈希标识"""
pattern = r'\((\d{6})\)'
match = re.search(pattern, filename)
if match:
return match.group(1)
else:
import hashlib
filename_clean = filename.replace(".xlsx", "").strip()
return hashlib.md5(filename_clean.encode()).hexdigest()[:6]
def get_all_excel_files(folder_path: str = ".") -> list:
"""获取当前文件夹下所有有效.xlsx文件（排除临时文件）"""
excel_files = [
os.path.join(folder_path, f)
for f in os.listdir(folder_path)
if f.endswith(".xlsx") and not f.startswith("~$")
]
if not excel_files:
print("⚠️ 未在当前文件夹找到Excel文件，程序退出！")
sys.exit()
return excel_files
# -------------------------------------------------
# 1. 数据读取（自动识别表头+清洗）
# -------------------------------------------------
def load_data(file_path: str) -> tuple[pd.DataFrame, str, str]:
filename = os.path.basename(file_path).replace(".xlsx", "")
stock_name = re.sub(r'\(\d{6}\)', '', filename).strip()
stock_code = extract_stock_code(filename)
print(f"\n🔍 正在处理 {filename}，自动探测表头...")
header_row = None
for test_row in range(5):
try:
temp_df = pd.read_excel(file_path, header=test_row, nrows=1, engine='openpyxl')
temp_cols = [str(col).strip() for col in temp_df.columns]
has_date = any("日期" in col for col in temp_cols)
has_close = any("收盘" in col for col in temp_cols)
if has_date and has_close:
header_row = test_row
print(f"✅ 自动识别表头行：第{header_row + 1}行（Excel中显示行号）")
break
except Exception:
continue
if header_row is None:
print(f"❌ {filename} 未找到有效表头，跳过该文件")
return None, None, None
try:
df = pd.read_excel(file_path, header=header_row, engine='openpyxl')
df.columns = [re.sub(r'[\s_\-]+', '', str(col)).strip() for col in df.columns]
print(f"📋 清洗后列名：{list(df.columns)}")
except Exception as e:
print(f"❌ 读取 {filename} 失败：{str(e)}，跳过")
return None, None, None
col_mapping = {}
date_candidates = [col for col in df.columns if "日期" in col]
for col in date_candidates:
try:
pd.to_datetime(df[col], errors="raise")
col_mapping["date"] = col
break
except Exception:
continue
if "date" not in col_mapping:
print(f"❌ {filename} 无有效日期列，跳过")
return None, None, None
close_candidates = [col for col in df.columns if "收盘" in col]
for col in close_candidates:
try:
pd.to_numeric(df[col], errors="raise")
col_mapping["close"] = col
break
except Exception:
continue
if "close" not in col_mapping:
print(f"❌ {filename} 无有效收盘价列，跳过")
return None, None, None
vol_candidates = [col for col in df.columns if any(k in col for k in ["成交", "量"])]
if vol_candidates:
for col in vol_candidates:
try:
pd.to_numeric(df[col], errors="raise")
col_mapping["vol"] = col
break
except Exception:
continue
col_mapping["vol"] = col_mapping.get("vol")
selected_cols = ["date", "close"] + (["vol"] if col_mapping["vol"] else [])
df_clean = df[[col_mapping[col] for col in selected_cols]].copy()
df_clean.columns = selected_cols
df_clean["date"] = pd.to_datetime(df_clean["date"], errors="coerce")
df_clean = df_clean.dropna(subset=["date", "close"]).set_index("date").sort_index()
if "vol" not in df_clean.columns:
df_clean["vol"] = 0
df_clean["has_vol"] = False
else:
df_clean["vol"] = pd.to_numeric(df_clean["vol"], errors="coerce").fillna(0)
df_clean["has_vol"] = (df_clean["vol"] > 0).any()
print(f"📊 数据预处理完成：{len(df_clean)} 条有效数据")
return df_clean, stock_name, stock_code
# -------------------------------------------------
# 2. 技术指标计算
# -------------------------------------------------
def calculate_tech_indicators(df: pd.DataFrame) -> pd.DataFrame:
df_copy = df.copy()
df_copy["slope10"] = (df_copy["close"] - df_copy["close"].shift(10)) / df_copy["close"].shift(10) * 100 / 10
df_copy["slope10_ok"] = df_copy["slope10"].between(0.5, 2)
df_copy["ma5"] = df_copy["close"].rolling(5).mean()
df_copy["ma10"] = df_copy["close"].rolling(10).mean()
df_copy["gold_cross"] = (df_copy["ma5"] > df_copy["ma10"]) & (df_copy["ma5"].shift(1) <= df_copy["ma10"].shift(1))
ema12 = df_copy["close"].ewm(span=12, adjust=False).mean()
ema26 = df_copy["close"].ewm(span=26, adjust=False).mean()
df_copy["dif"] = ema12 - ema26
df_copy["dea"] = df_copy["dif"].ewm(span=9, adjust=False).mean()
df_copy["macd"] = 2 * (df_copy["dif"] - df_copy["dea"])
df_copy["macd_inc"] = (df_copy["macd"] > 0) & (df_copy["macd"] > df_copy["macd"].shift(1))
if df_copy["has_vol"].iloc[0]:
df_copy["ma5vol"] = df_copy["vol"].rolling(5).mean()
df_copy["vol_ratio_ok"] = df_copy.apply(
lambda x: 1.2 <= (x["vol"] / x["ma5vol"]) <= 1.5 if x["ma5vol"] > 0 else False, axis=1
)
else:
df_copy["vol_ratio_ok"] = True
df_copy["buy_signal"] = df_copy["slope10_ok"] & df_copy["gold_cross"] & df_copy["macd_inc"] & df_copy[
"vol_ratio_ok"]
return df_copy
# -------------------------------------------------
# 3. 策略回测
# -------------------------------------------------
def backtest_strategy(df: pd.DataFrame, stock_code: str, init_cash: float = 1_000_000) -> tuple[pd.DataFrame, float]:
cash = init_cash
position = {}
trade_log = []
for dt, row in df.iterrows():
current_close = row["close"]
current_slope = row["slope10"]
if position:
code = list(position.keys())[0]
pos = position[code]
buy_price = pos["buy_price"]
shares = pos["shares"]
profit_rate = (current_close - buy_price) / buy_price
if profit_rate >= 0.15:
sell_amount = current_close * shares
cash += sell_amount
trade_log.append(
[dt.strftime("%Y-%m-%d"), code, "止盈全平", round(current_close, 2), shares, round(sell_amount, 2)])
del position[code]
elif 0.08 <= profit_rate < 0.15:
sell_shares = shares // 2
sell_amount = current_close * sell_shares
cash += sell_amount
trade_log.append([dt.strftime("%Y-%m-%d"), code, "止盈半平", round(current_close, 2), sell_shares,
round(sell_amount, 2)])
position[code]["shares"] -= sell_shares
stop_loss_rate = 0.03 if (0.5 <= pos["buy_slope"] < 1) else 0.02
if current_close <= buy_price * (1 - stop_loss_rate):
sell_amount = current_close * shares
cash += sell_amount
trade_log.append(
[dt.strftime("%Y-%m-%d"), code, "止损", round(current_close, 2), shares, round(sell_amount, 2)])
del position[code]
if row["buy_signal"] and not position:
max_buy_amount = min(cash * 0.05, cash * 0.15)
if max_buy_amount < current_close:
continue
buy_shares = int(max_buy_amount / current_close)
buy_amount = current_close * buy_shares
cash -= buy_amount
position[stock_code] = {"shares": buy_shares, "buy_price": current_close, "buy_slope": current_slope}
trade_log.append([dt.strftime("%Y-%m-%d"), stock_code, "买入", round(current_close, 2), buy_shares,
round(buy_amount, 2)])
if position:
code = list(position.keys())[0]
pos = position[code]
sell_amount = df["close"].iloc[-1] * pos["shares"]
cash += sell_amount
trade_log.append(
[df.index[-1].strftime("%Y-%m-%d"), code, "期末清仓", round(df["close"].iloc[-1], 2), pos["shares"],
round(sell_amount, 2)])
log_cols = ["日期", "证券代码", "操作类型", "价格(元)", "股数", "金额(元)"]
log_df = pd.DataFrame(trade_log, columns=log_cols) if trade_log else pd.DataFrame(columns=log_cols)
return log_df, round(cash, 2)
# -------------------------------------------------
# 4. 图表生成（修复黑色块核心优化）
# -------------------------------------------------
def plot_quant_analysis(df: pd.DataFrame, stock_name: str, stock_code: str) -> BytesIO:
"""生成4合一分析图表，返回内存流（修复背景与渲染）"""
# 创建图表对象并设置背景
fig = plt.figure(figsize=(16, 12), facecolor='white', edgecolor='none')
fig.suptitle(f"{stock_name}({stock_code}) 量化分析图",
fontsize=16, fontweight="bold", y=0.98)
# 创建子图
ax1 = fig.add_subplot(2, 2, 1)
ax2 = fig.add_subplot(2, 2, 2)
ax3 = fig.add_subplot(2, 2, 3)
ax4 = fig.add_subplot(2, 2, 4)
# 子图1：价格+买入信号+均线
ax1.plot(df.index, df["close"], color="#1f77b4", linewidth=1.5, label="收盘价")
buy_points = df[df["buy_signal"]]
if not buy_points.empty:
ax1.scatter(buy_points.index, buy_points["close"],
color="red", s=80, marker="o", edgecolors="white", linewidth=1.5, label="买入信号")
ax1.plot(df.index, df["ma5"], color="#ff7f0e", linewidth=1, label="5日均线")
ax1.plot(df.index, df["ma10"], color="#2ca02c", linewidth=1, label="10日均线")
ax1.set_title("价格走势与买入信号", fontsize=12)
ax1.set_ylabel("价格(元)")
ax1.legend(fontsize=10)
ax1.grid(alpha=0.3)
ax1.set_facecolor('white') # 子图背景强制白色
# 子图2：10日斜率
ax2.plot(df.index, df["slope10"], color="#d62728", linewidth=1.5, label="10日斜率")
ax2.axhline(y=0.5, color="orange", linestyle="--", alpha=0.7, label="最佳下界(0.5%)")
ax2.axhline(y=2, color="orange", linestyle="--", alpha=0.7, label="最佳上界(2%)")
ax2.fill_between(df.index, 0.5, 2, color="orange", alpha=0.1, label="最佳中速区间")
ax2.set_title("10日斜率变化", fontsize=12)
ax2.set_ylabel("斜率(%)")
ax2.legend(fontsize=10)
ax2.grid(alpha=0.3)
ax2.set_facecolor('white') # 子图背景强制白色
# 子图3：MACD
ax3.plot(df.index, df["dif"], color="#1f77b4", linewidth=1.5, label="DIF")
ax3.plot(df.index, df["dea"], color="#ff7f0e", linewidth=1.5, label="DEA")
macd_pos = df[df["macd"] > 0]
macd_neg = df[df["macd"] <= 0]
ax3.bar(macd_pos.index, macd_pos["macd"], color="red", alpha=0.5, label="MACD红柱")
ax3.bar(macd_neg.index, macd_neg["macd"], color="green", alpha=0.5, label="MACD绿柱")
ax3.set_title("MACD指标", fontsize=12)
ax3.set_ylabel("MACD值")
ax3.legend(fontsize=10)
ax3.grid(alpha=0.3)
ax3.set_facecolor('white') # 子图背景强制白色
# 子图4：成交量（修复无数据时的黑色块）
ax4.set_facecolor('white') # 子图背景强制白色
if df["has_vol"].iloc[0]:
ax4.bar(df.index, df["vol"], color="#2ca02c", alpha=0.6, label="成交量")
ax4.set_title("成交量变化", fontsize=12)
ax4.set_ylabel("成交量")
ax4.legend(fontsize=10)
else:
# 无数据时，用白色背景+黑色文本，避免黑色块
ax4.text(0.5, 0.5, "无成交量数据",
ha="center", va="center", transform=ax4.transAxes,
fontsize=12, color="black", fontweight="bold")
ax4.set_title("成交量变化", fontsize=12)
ax4.grid(alpha=0.3)
# 优化渲染：避免边缘裁剪
plt.tight_layout()
plt.subplots_adjust(top=0.93)
# 生成图像流：降低dpi避免流过大，确保完整
img_stream = BytesIO()
fig.savefig(
img_stream,
dpi=150, # 降低dpi减少内存占用，避免流损坏
bbox_inches="tight",
format="png",
facecolor='white', # 再次确认背景色
edgecolor='none', # 去除图像边缘
transparent=False # 禁用透明
)
img_stream.seek(0) # 重置流指针
plt.close(fig) # 关闭图表释放内存
print(f"📊 图表生成成功（内存流大小：{img_stream.getbuffer().nbytes / 1024:.2f} KB）")
return img_stream
# -------------------------------------------------
# 5. 生成PDF子模块（修复图像显示）
# -------------------------------------------------
def generate_sub_pdf(df_indicators: pd.DataFrame, log_df: pd.DataFrame,
stock_name: str, stock_code: str, final_cash: float, font_name: str) -> str:
"""生成单股票PDF子模块"""
current_dir = os.getcwd()
sub_pdf_path = os.path.join(current_dir, f"temp_{stock_code}_sub.pdf")
styles = getSampleStyleSheet()
# 创建自定义样式确保字体正确应用（修复小标题字体问题）
custom_heading1 = ParagraphStyle(
name='CustomHeading1',
parent=styles['Heading1'],
fontName=font_name,
fontSize=16,
spaceAfter=15
)
custom_heading2 = ParagraphStyle(
name='CustomHeading2',
parent=styles['Heading2'],
fontName=font_name,
fontSize=12,
spaceAfter=6
)
custom_normal = ParagraphStyle(
name='CustomNormal',
parent=styles['Normal'],
fontName=font_name,
fontSize=10,
spaceAfter=6
)
elements = []
# 子模块标题
title = Paragraph(f"<b>{stock_name}({stock_code}) 量化分析子模块</b>", custom_heading1)
title.alignment = 1
elements.append(title)
elements.append(Spacer(1, 10))
# 交易记录（修复表格字体）
elements.append(Paragraph("<b>1. 交易记录</b>", custom_heading2))
elements.append(Spacer(1, 8))
if log_df.empty:
elements.append(Paragraph("⚠️ 无符合条件的买入信号，未产生交易记录", custom_normal))
else:
table_data = [log_df.columns.tolist()]
for _, row in log_df.iterrows():
row_data = row.tolist()
row_data[3] = round(row_data[3], 2)
row_data[5] = round(row_data[5], 2)
table_data.append(row_data)
col_widths = [100, 100, 120, 100, 80, 120]
table = Table(table_data, colWidths=col_widths, repeatRows=1)
# 修复表格样式：所有单元格应用字体，避免黑色块
table_style = TableStyle([
("BACKGROUND", (0, 0), (-1, 0), colors.lightblue),
("TEXTCOLOR", (0, 0), (-1, -1), colors.black), # 所有文本强制黑色
("ALIGN", (0, 0), (-1, -1), "CENTER"),
("FONTNAME", (0, 0), (-1, -1), font_name), # 所有单元格用指定字体
("FONTSIZE", (0, 0), (-1, -1), 10),
("FONTWEIGHT", (0, 0), (-1, 0), "BOLD"),
("GRID", (0, 0), (-1, -1), 1, colors.black),
("ROWBACKGROUNDS", (0, 1), (-1, -1), [colors.white, colors.lightgrey]),
])
table.setStyle(table_style)
elements.append(table)
elements.append(Spacer(1, 20))
# 量化分析图表（修复图像尺寸）
elements.append(Paragraph("<b>2. 量化分析图表</b>", custom_heading2))
elements.append(Spacer(1, 8))
try:
img_stream = plot_quant_analysis(df_indicators, stock_name, stock_code)
img_data = img_stream.getvalue() # 直接获取二进制数据
# 使用ReportLab的ImageReader处理内存数据
img_reader = ImageReader(BytesIO(img_data))
# 计算合适的图像尺寸（A4页面宽度减去边距）
page_width = A4[0] - 60 # 左右各30mm边距
img_width = page_width
img_height = img_width * 0.75 # 保持4:3比例
img = Image(
img_reader,
width=img_width,
height=img_height,
kind='direct' # 直接嵌入图像数据
)
img.hAlign = "CENTER"
elements.append(img)
except Exception as e:
error_msg = f"⚠️ 图表生成失败：{str(e)}"
elements.append(Paragraph(error_msg, custom_normal))
print(f"❌ {error_msg}")
finally:
if 'img_stream' in locals():
img_stream.close() # 确保关闭内存流
elements.append(Spacer(1, 20))
# 子模块小结
elements.append(Paragraph("<b>3. 子模块小结</b>", custom_heading2))
elements.append(Spacer(1, 8))
buy_count = df_indicators["buy_signal"].sum()
init_cash = 1_000_000
profit = final_cash - init_cash
profit_rate = (profit / init_cash) * 100
summary_texts = [
f"• 识别买入信号数：{buy_count} 个",
f"• 初始资金：{init_cash:,.2f} 元",
f"• 最终资金：{final_cash:,.2f} 元",
f"• 绝对收益：{profit:,.2f} 元（{profit_rate:.2f}%）",
f"• 数据时间范围：{df_indicators.index.min().strftime('%Y-%m-%d')} ~ {df_indicators.index.max().strftime('%Y-%m-%d')}"
]
for text in summary_texts:
elements.append(Paragraph(text, custom_normal))
# 生成子PDF
try:
doc = SimpleDocTemplate(
sub_pdf_path,
pagesize=A4,
rightMargin=30,
leftMargin=30,
topMargin=30,
bottomMargin=20,
defaultFontName=font_name # 设置默认字体
)
doc.build(elements)
print(f"✅ 生成 {stock_name} 子模块PDF：{os.path.basename(sub_pdf_path)}")
return sub_pdf_path
except Exception as e:
print(f"❌ 子PDF生成失败：{str(e)}")
raise
# -------------------------------------------------
# 6. 合并总PDF报告（修复合并问题）
# -------------------------------------------------
def merge_to_total_pdf(sub_pdf_paths: list, stock_info_list: list, font_name: str,
total_pdf_name: str = "多股票量化策略总报告.pdf") -> None:
"""合并所有子PDF为总报告"""
current_dir = os.getcwd()
total_pdf_path = os.path.join(current_dir, total_pdf_name)
cover_dir_pdf = os.path.join(current_dir, "temp_cover_dir.pdf")
styles = getSampleStyleSheet()
# 创建自定义样式
custom_heading1 = ParagraphStyle(
name='CustomHeading1',
parent=styles['Heading1'],
fontName=font_name,
fontSize=18,
spaceAfter=20
)
custom_heading2 = ParagraphStyle(
name='CustomHeading2',
parent=styles['Heading2'],
fontName=font_name,
fontSize=14,
spaceAfter=12
)
custom_normal = ParagraphStyle(
name='CustomNormal',
parent=styles['Normal'],
fontName=font_name,
fontSize=11,
spaceAfter=8
)
elements = []
# 封面
cover_title = Paragraph("<b>多股票中速上升段量化策略总报告</b>", custom_heading1)
cover_title.alignment = 1
elements.append(cover_title)
elements.append(Spacer(1, 30))
report_date = Paragraph(f"报告生成时间：{datetime.now().strftime('%Y-%m-%d %H:%M:%S')}", custom_normal)
report_date.alignment = 1
elements.append(report_date)
elements.append(Spacer(1, 10))
stock_count = Paragraph(f"分析股票数量：{len(stock_info_list)} 只", custom_normal)
stock_count.alignment = 1
elements.append(stock_count)
elements.append(Spacer(1, 60))
# 股票列表（修复表格字体）
elements.append(Paragraph("<b>参与分析的股票列表</b>", custom_heading2))
elements.append(Spacer(1, 8))
stock_table_data = [["股票名称", "证券代码"]] + stock_info_list
table = Table(stock_table_data, colWidths=[250, 100], repeatRows=1)
table.setStyle(TableStyle([
("BACKGROUND", (0, 0), (-1, 0), colors.lightblue),
("TEXTCOLOR", (0, 0), (-1, -1), colors.black),
("ALIGN", (0, 0), (-1, -1), "CENTER"),
("FONTNAME", (0, 0), (-1, -1), font_name), # 所有单元格用指定字体
("FONTSIZE", (0, 0), (-1, -1), 10),
("FONTWEIGHT", (0, 0), (-1, 0), "BOLD"),
("GRID", (0, 0), (-1, -1), 1, colors.black),
("ROWBACKGROUNDS", (0, 1), (-1, -1), [colors.white, colors.lightgrey]),
]))
elements.append(table)
elements.append(Spacer(1, 60))
# 总目录
elements.append(Paragraph("<b>总目录</b>", custom_heading2))
elements.append(Spacer(1, 8))
elements.append(Paragraph("1. 报告说明与风险提示", custom_normal))
for i, (stock_name, stock_code) in enumerate(stock_info_list, 2):
elements.append(Paragraph(f"{i}. {stock_name}({stock_code}) 量化分析子模块", custom_normal))
elements.append(Spacer(1, 40))
# 风险提示
elements.append(Paragraph("<b>1. 报告说明与风险提示</b>", custom_heading2))
elements.append(Spacer(1, 8))
note_texts = [
"• 本报告基于历史数据回测，不构成投资建议。",
"• 策略核心：10日斜率0.5%-2%+均线金叉+MACD红柱放大+量比1.2-1.5。",
"• 仓位规则：单票首仓≤5%，严格止盈止损（8%-15%止盈，2%-3%止损）。",
"• 风险提示：市场波动可能导致策略失效，投资需谨慎。"
]
for text in note_texts:
elements.append(Paragraph(text, custom_normal))
# 生成封面+目录PDF
doc = SimpleDocTemplate(
cover_dir_pdf,
pagesize=A4,
rightMargin=30,
leftMargin=30,
topMargin=30,
bottomMargin=20,
defaultFontName=font_name # 设置默认字体
)
doc.build(elements)
print(f"\n📄 封面+目录PDF生成完成")
# 合并PDF前验证子PDF有效性
valid_sub_pdfs = []
for sub_pdf in sub_pdf_paths:
if os.path.exists(sub_pdf) and os.path.getsize(sub_pdf) > 1024: # 排除空文件
valid_sub_pdfs.append(sub_pdf)
print(f"🔗 待合并子PDF：{os.path.basename(sub_pdf)}（大小：{os.path.getsize(sub_pdf) / 1024:.2f} KB）")
else:
print(f"❌ 跳过损坏的子PDF：{os.path.basename(sub_pdf)}")
# 合并PDF
merger = PdfMerger()
merger.append(cover_dir_pdf)
for sub_pdf in valid_sub_pdfs:
merger.append(sub_pdf)
print(f"✅ 合并子PDF：{os.path.basename(sub_pdf)}")
merger.write(total_pdf_path)
merger.close()
print(f"\n🎉 总报告生成完成：{total_pdf_path}")
# 确认总PDF正常后删除临时文件
if os.path.exists(total_pdf_path) and os.path.getsize(total_pdf_path) > 1024:
os.remove(cover_dir_pdf)
for sub_pdf in valid_sub_pdfs:
os.remove(sub_pdf)
print("🗑️ 临时文件已删除")
else:
print("⚠️ 总PDF生成异常，保留临时文件以便排查")
# -------------------------------------------------
# 主流程
# -------------------------------------------------
def main():
print("=" * 70)
print("多股票量化策略总报告生成器（修复版）")
print("=" * 70)
# 1. 初始化中文字体
font_name = init_chinese_font()
# 2. 初始化Matplotlib字体
init_matplotlib_font(font_name)
# 3. 获取Excel文件
excel_files = get_all_excel_files()
print(f"\n📂 找到 {len(excel_files)} 个Excel文件，处理顺序：")
for i, file in enumerate(excel_files, 1):
print(f" {i}. {os.path.basename(file)}")
# 4. 批量处理
sub_pdf_paths = []
stock_info_list = []
init_cash = 1_000_000
for file in excel_files:
print(f"\n" + "-" * 50)
print(f"开始处理：{os.path.basename(file)}")
print("-" * 50)
df_raw, stock_name, stock_code = load_data(file)
if df_raw is None:
continue
df_indicators = calculate_tech_indicators(df_raw)
print(f"📈 技术指标计算完成：买入信号数 = {df_indicators['buy_signal'].sum()} 个")
log_df, final_cash = backtest_strategy(df_indicators, stock_code, init_cash)
print(f"📊 回测完成：最终资金 = {final_cash:,.2f} 元")
sub_pdf = generate_sub_pdf(df_indicators, log_df, stock_name, stock_code, final_cash, font_name)
sub_pdf_paths.append(sub_pdf)
stock_info_list.append((stock_name, stock_code))
# 5. 合并总报告
if sub_pdf_paths:
merge_to_total_pdf(sub_pdf_paths, stock_info_list, font_name)
else:
print("\n⚠️ 无有效子PDF可合并，无法生成总报告！")
print("\n" + "=" * 70)
print("所有流程执行完毕！")
print("=" * 70)
# =========================================================
# Towin_robot.py 增量补丁（红♥历史信号 + 蓝♥未来节点 + 最优价）
# 使用方式：直接粘到原文件尾部，原重复定义已剔除
# =========================================================
# ---------------- 新增 import ----------------
import matplotlib.dates as mdates
from reportlab.platypus import PageBreak
# ---------------------------------------------------------
# 1. 扩展指标：计算最优买入价 & 未来时间窗口
# ---------------------------------------------------------
def enrich_future_nodes(df: pd.DataFrame) -> pd.DataFrame:
"""
1) 为历史信号日计算“最优买入价”
优先用 (开盘+最低)/2，缺开盘则用 close*0.98
2) 外推 2025-12-31 前所有「未来潜在节点」
规则：中速 0.5–2 %/日 或 高速 >2 %/日
"""
# ---- 最优价 ----
if "open" not in df.columns:
df["open"] = df["close"].shift(1).fillna(df["close"] * 0.99)
if "low" not in df.columns:
df["low"] = df["close"] * 0.97
df["optimal_buy"] = (df["open"] + df["low"]) / 2
# ---- 未来节点 ----
last_day = df.index[-1]
future_end = pd.Timestamp("2025-12-31")
holiday_keys = pd.date_range(last_day + pd.Timedelta(days=1), future_end, freq="B")
latest_slope = df["slope10"].iloc[-1] / 100
base_price = df["close"].iloc[-1]
future_prices = [base_price * (1 + latest_slope) ** i for i in range(1, len(holiday_keys) + 1)]
future_df = pd.DataFrame(index=holiday_keys, data={"close": future_prices})
future_df["slope10"] = (future_df["close"] - future_df["close"].shift(10)) / future_df["close"].shift(10) * 100 / 10
future_df["future_node"] = False
future_df.loc[future_df["slope10"].between(0.5, 2), "future_node"] = True
future_df.loc[future_df["slope10"] > 2, "future_node"] = True
future_df["optimal_buy"] = future_df["close"] * 0.98
df_future = future_df[future_df["future_node"]].copy()
# 合并回主表（仅用于绘图）
df["is_future"] = False
for dt in df_future.index:
df.loc[dt, "is_future"] = True
df.loc[dt, "optimal_buy"] = df_future.loc[dt, "optimal_buy"]
return df, df_future
# ---------------------------------------------------------
# 2. 绘图：红♥历史 + 蓝♥未来 + 最优价标注
# ---------------------------------------------------------
def plot_quant_analysis_plus(df: pd.DataFrame, stock_name: str, stock_code: str):
"""返回内存图 + 未来节点df"""
df, df_future = enrich_future_nodes(df)
fig = plt.figure(figsize=(16, 14), facecolor='white')
fig.suptitle(f"{stock_name}({stock_code}) 量化分析图（红♥历史信号 / 蓝♥未来节点）", fontsize=16, y=0.98)
ax1 = plt.subplot(3, 2, 1)
ax2 = plt.subplot(3, 2, 2)
ax3 = plt.subplot(3, 2, 3)
ax4 = plt.subplot(3, 2, 4)
ax5 = plt.subplot(3, 1, 3)
# 1) 价格+红♥
ax1.plot(df.index, df["close"], color="#1f77b4", lw=1.5, label="收盘价")
hist_buy = df[df["buy_signal"].fillna(False)]
if not hist_buy.empty:
ax1.scatter(hist_buy.index, hist_buy["close"], marker="$♥$", s=120, color="red", label="历史买入信号", zorder=5)
ax1.plot(df.index, df["ma5"], "#ff7f0e", lw=1, label="5日均线")
ax1.plot(df.index, df["ma10"], "#2ca02c", lw=1, label="10日均线")
ax1.set_title("收盘价与历史买入信号（红♥）"); ax1.legend(); ax1.grid(alpha=0.3)
# 2) 斜率
ax2.plot(df.index, df["slope10"], color="#d62728", lw=1.5)
ax2.axhline(0.5, ls="--", c="orange"); ax2.axhline(2, ls="--", c="orange")
ax2.fill_between(df.index, 0.5, 2, color="orange", alpha=0.1)
ax2.set_title("10 日斜率（中速 0.5–2 %/日）"); ax2.grid(alpha=0.3)
# 3) MACD
ax3.plot(df.index, df["dif"], "#1f77b4", label="DIF")
ax3.plot(df.index, df["dea"], "#ff7f0e", label="DEA")
macd_pos = (df["macd"] > 0).fillna(False)
ax3.bar(df.index[macd_pos], df["macd"][macd_pos], color="red", alpha=0.5, label="红柱")
ax3.bar(df.index[~macd_pos], df["macd"][~macd_pos], color="green", alpha=0.5, label="绿柱")
ax3.set_title("MACD"); ax3.legend(); ax3.grid(alpha=0.3)
# 4) 成交量
ax4.bar(df.index, df["vol"], color="#2ca02c", alpha=0.6)
ax4.set_title("成交量"); ax4.grid(alpha=0.3)
# 5) 未来节点（蓝♥）
ax5.plot(df.index, df["close"], color="gray", lw=1, alpha=0.7, label="历史收盘")
if not hist_buy.empty:
ax5.scatter(hist_buy.index, hist_buy["close"], marker="$♥$", s=120, color="red", label="历史买入节点", zorder=6)
if not df_future.empty:
ax5.scatter(df_future.index, df_future["close"], marker="$♥$", s=120, color="blue", label="未来潜在节点", zorder=6)
# 标注最优价（前 10 个）
for dt, row in df_future.head(10).iterrows():
ax5.text(dt, row["close"] * 0.95, f"{row['optimal_buy']:.2f}", ha="center", va="top",
fontsize=8, color="blue", alpha=0.9)
ax5.set_title("未来买入节点预测（蓝♥，含最优价）")
ax5.legend(); ax5.grid(alpha=0.3)
ax5.xaxis.set_major_formatter(mdates.DateFormatter('%Y-%m'))
fig.autofmt_xdate()
# 内存流
img_stream = BytesIO()
fig.savefig(img_stream, dpi=150, bbox_inches="tight", facecolor='white', format="png")
img_stream.seek(0)
plt.close(fig)
return img_stream, df_future
# ---------------------------------------------------------
# 3. 生成子 PDF（新增“未来节点”一页）
# ---------------------------------------------------------
def generate_sub_pdf_plus(df_indicators: pd.DataFrame, log_df: pd.DataFrame,
stock_name: str, stock_code: str, final_cash: float, font_name: str) -> str:
current_dir = os.getcwd()
sub_pdf_path = os.path.join(current_dir, f"temp_{stock_code}_sub.pdf")
styles = getSampleStyleSheet()
for st in [styles['Heading1'], styles['Heading2'], styles['Normal']]:
st.fontName = font_name
elements = []
# 1) 封面/交易记录/图表
title = Paragraph(f"<b>{stock_name}({stock_code}) 量化分析子模块</b>", styles['Heading1'])
title.alignment = 1; elements.extend([title, Spacer(1, 10)])
elements.append(Paragraph("<b>1. 交易记录</b>", styles['Heading2']))
if log_df.empty:
elements.append(Paragraph("⚠️ 无交易记录", styles['Normal']))
else:
table_data = [log_df.columns.tolist()] + log_df.round(2).values.tolist()
tbl = Table(table_data, colWidths=[100, 100, 120, 100, 80, 120], repeatRows=1)
tbl.setStyle(TableStyle([
("BACKGROUND", (0, 0), (-1, 0), colors.lightblue),
("TEXTCOLOR", (0, 0), (-1, -1), colors.black),
("ALIGN", (0, 0), (-1, -1), "CENTER"),
("FONTNAME", (0, 0), (-1, -1), font_name),
("GRID", (0, 0), (-1, -1), 1, colors.black),
]))
elements.append(tbl)
elements.append(Spacer(1, 20))
# 2) 图表（红蓝双心）
elements.append(Paragraph("<b>2. 量化分析图表（红♥历史 / 蓝♥未来）</b>", styles['Heading2']))
img_stream, df_future = plot_quant_analysis_plus(df_indicators, stock_name, stock_code)
img = Image(img_stream, width=A4[0] - 60, height=(A4[0] - 60) * 0.75)
img.hAlign = "CENTER"
elements.append(img)
elements.append(PageBreak())
# 3) 未来节点
elements.append(Paragraph("<b>3. 未来潜在买入节点（外推至 2025-12-31）</b>", styles['Heading2']))
if df_future.empty:
elements.append(Paragraph("• 未探测到满足中/高速区间的未来节点。", styles['Normal']))
else:
future_tbl_data = [["日期", "预估收盘价(元)", "最优买入价(元)", "斜率(%)"]] + \
[[d.strftime("%Y-%m-%d"), f"{r['close']:.2f}", f"{r['optimal_buy']:.2f}", f"{r['slope10']:.2f}"]
for d, r in df_future.head(20).iterrows()]
ftbl = Table(future_tbl_data, colWidths=[120, 120, 120, 80], repeatRows=1)
ftbl.setStyle(TableStyle([
("BACKGROUND", (0, 0), (-1, 0), colors.lightgreen),
("TEXTCOLOR", (0, 0), (-1, -1), colors.black),
("ALIGN", (0, 0), (-1, -1), "CENTER"),
("FONTNAME", (0, 0), (-1, -1), font_name),
("GRID", (0, 0), (-1, -1), 1, colors.black),
]))
elements.append(ftbl)
elements.append(Spacer(1, 10))
elements.append(Paragraph(f"• 共探测到 <b>{len(df_future)}</b> 个未来节点，上表展示前 20 个。", styles['Normal']))
# ========== 第七步：胜率 & 蒙特卡洛可信度 & 自动重训 ==========
# 1) 计算胜率
win_rate, sig_cnt, avg_ret = calc_win_rate(df_indicators)
model_auc = auto_retrain(df_indicators)
# 2) 蒙特卡洛外推可信度节点
mask_days = get_trade_days(df_indicators.index[-1].strftime('%Y%m%d'), '20251231')
mc_df = monte_carlo_nodes(
df_indicators['close'].iloc[-1],
df_indicators['close'].pct_change().mean() * 252,
df_indicators['close'].pct_change().std() * np.sqrt(252),
df_indicators.index[-1],
'20251231',
mask_days
)
# 3) 写进 PDF
elements.append(Spacer(1, 12))
elements.append(Paragraph("<b>5. 胜率与蒙特卡洛可信度</b>", styles['Heading2']))
if win_rate is None:
elements.append(Paragraph("• 信号样本不足，暂无法计算胜率。", styles['Normal']))
else:
elements.append(Paragraph(
f"• 历史胜率：<b>{win_rate:.1%}</b> 信号次数：<b>{sig_cnt}</b> 平均收益：<b>{avg_ret:.2%}</b>",
styles['Normal']))
if mc_df.empty:
elements.append(Paragraph("• 蒙特卡洛外推：未探测到可信度 ≥80% 节点。", styles['Normal']))
else:
mc_tbl_data = [["日期", "预估价(元)", "可信度"]] + \
[[d.strftime('%Y-%m-%d'), f"{p:.2f}", f"{c:.0%}"] for d, p, c in mc_df.values]
mc_tbl = Table(mc_tbl_data, colWidths=[120, 100, 80], repeatRows=1)
mc_tbl.setStyle(TableStyle([
("BACKGROUND", (0, 0), (-1, 0), colors.lightgreen),
("TEXTCOLOR", (0, 0), (-1, -1), colors.black),
("ALIGN", (0, 0), (-1, -1), "CENTER"),
("FONTNAME", (0, 0), (-1, -1), font_name),
("GRID", (0, 0), (-1, -1), 1, colors.black),
]))
elements.append(mc_tbl)
# 4) 推送钩子
push_lark(f"{stock_name}({stock_code}) "
f"胜率={(win_rate or 0):.1%} "
f"信号={sig_cnt or 0} "
f"可信度节点={len(mc_df)}")
# ========== 第七步结束 ==========
# 4) 小结
elements.append(Spacer(1, 20))
elements.append(Paragraph("<b>4. 子模块小结</b>", styles['Heading2']))
buy_cnt = df_indicators["buy_signal"].sum()
profit = final_cash - 1_000_000
profit_rate = profit / 1_000_000 * 100
summary = f"""
• 历史买入信号：<b>{buy_cnt}</b> 个<br/>
• 最终资金：<b>{final_cash:,.2f}</b> 元（收益 <b>{profit_rate:.2f}%</b>）<br/>
• 未来节点：<b>{len(df_future)}</b> 个（至 2025-12-31）<br/>
• 数据范围：{df_indicators.index.min().strftime('%Y-%m-%d')} ~ {df_indicators.index.max().strftime('%Y-%m-%d')}
"""
elements.append(Paragraph(summary, styles['Normal']))
# 5) 生成
SimpleDocTemplate(sub_pdf_path, pagesize=A4, rightMargin=30, leftMargin=30, topMargin=30, bottomMargin=20,
defaultFontName=font_name).build(elements)
print(f"✅ 生成 {stock_name} 子模块PDF（含未来节点）：{os.path.basename(sub_pdf_path)}")
return sub_pdf_path
# ---------- 自动重训 & 监控参数 ----------
ROLL_WINDOW = 240 # 滚动 240 日训练集
RETRAIN_WEEKDAY = 4 # 4=周五
MIN_SIGNAL_COUNT = 20 # 最少信号样本
WIN_THRESHOLD = 0.08 # 8% 以上视为盈利
IC_DECAY_ALERT = 0.30 # IC 衰减 30% 触发警告
MONTE_PATHS = 5000 # 蒙特卡洛路径
MONTE_CONF = 0.80 # 可信度置信水平
WEBHOOK_URL = "" # 飞书机器人 webhook，留空则不推送
#新增工具函数
def get_trade_days(start, end):
"""本地交易日历（周一到周五，剔除法定节假日简单列表）"""
# 2024-2026 主要节假日（可再自行补充）
holidays = [
'20241001', '20241002', '20241003', '20241004', '20241007', # 国庆
'20250203', '20250204', '20250205', '20250206', '20250207', # 春节
'20250501', '20250502', '20250609', '20250610', # 劳动、端午
'20251001', '20251002', '20251003', '20251006', '20251007', # 2025 国庆
'20260101', '20260216', '20260217', '20260218', '20260219', '20260220' # 2026 春节
]
# 生成自然工作日
bdays = pd.bdate_range(start, end)
# 剔除节假日
hdays = pd.to_datetime(holidays)
return bdays.drop(bdays.intersection(hdays))
def calc_win_rate(df_raw, signal_col='buy_signal', ret_col='fwd_ret'):
"""计算胜率百分比"""
df = df_raw.copy()
# 未来 5 日收益
df[ret_col] = df['close'].shift(-5) / df['close'] - 1
sig = df[signal_col].fillna(False)
if sig.sum() < MIN_SIGNAL_COUNT:
return None, None, None
win = (df[ret_col] > WIN_THRESHOLD) & sig
win_rate = win.sum() / sig.sum()
return win_rate, sig.sum(), df[ret_col][sig].mean()
def ic_decay(df, factor, ret):
"""因子 IC 及环比衰减"""
df = df.copy()
df['factor'] = factor
df['ret'] = ret
ic = df[['factor', 'ret']].corr().iloc[0, 1]
ic_prev = df[['factor', 'ret']].shift(60).corr().iloc[0, 1]
decay = abs(ic - ic_prev) / (abs(ic_prev) + 1e-6)
return ic, decay
#-------------蒙特卡洛外推-----------------------------------------------------
def monte_carlo_nodes(last_price, mu, sigma, last_day, end_day, mask_days):
"""返回可信度节点 DataFrame"""
days = len(mask_days)
dt = 1/252
paths = np.exp((mu - 0.5*sigma**2)*dt + sigma*np.sqrt(dt)*np.random.normal(0, 1, (MONTE_PATHS, days)))
prices = last_price * np.cumprod(paths, axis=1)
slopes = (prices[:, 10:] - prices[:, :-10]) / prices[:, :-10] / 10 * 100
# 判断满足斜率区间
hit = ((slopes >= 0.5) & (slopes <= 2)) | (slopes > 2)
# 计算首次满足日期可信度
first_hit = np.full(MONTE_PATHS, np.nan)
for i in range(MONTE_PATHS):
hh = np.where(hit[i])[0]
if len(hh) > 0:
first_hit[i] = hh[0]
conf50 = np.nanpercentile(first_hit, 50)
conf80 = np.nanpercentile(first_hit, (1-MONTE_CONF)*100)
# 映射回日历
if not np.isnan(conf50):
date50 = mask_days[int(conf50)]
date80 = mask_days[int(conf80)]
price50 = last_price * (1 + mu)**int(conf50)
price80 = last_price * (1 + mu)**int(conf80)
return pd.DataFrame({'date':[date50, date80],
'price':[price50, price80],
'credib':[0.50, MONTE_CONF]})
else:
return pd.DataFrame()
# ------------------自动重训函数---------------------------------------
def auto_retrain(df):
"""滚动重训逻辑回归权重"""
today = pd.Timestamp.today()
if today.weekday() != RETRAIN_WEEKDAY:
return None
train = df.tail(ROLL_WINDOW).copy()
train['ret5'] = train['close'].shift(-5) / train['close'] - 1
train = train.dropna()
if len(train) < MIN_SIGNAL_COUNT:
return None
X = train[['slope10', 'macd', 'vol_ratio']].fillna(0)
y = (train['ret5'] > WIN_THRESHOLD).astype(int)
if y.sum() < MIN_SIGNAL_COUNT // 2:
return None
model = LogisticRegression().fit(X, y)
auc = roc_auc_score(y, model.predict_proba(X)[:, 1])
joblib.dump(model, 'signal_model.pkl')
print(f"[RETRAIN] {today.date()} 模型重训完成 AUC={auc:.3f}")
return auc
#----------推送钩子----------------------------------------
def push_lark(msg):
if not WEBHOOK_URL:
return
requests.post(WEBHOOK_URL, json={"msg_type":"text","content":{"text":msg}})
# 4. 替换原主流程入口
# ---------------------------------------------------------
def main_plus():
"""与原 main() 相同，仅替换子 PDF 生成函数"""
font_name = init_chinese_font()
init_matplotlib_font(font_name)
excel_files = get_all_excel_files()
sub_pdf_paths, stock_info_list = [], []
for file in excel_files:
df_raw, stock_name, stock_code = load_data(file)
if df_raw is None:
continue
df_indicators = calculate_tech_indicators(df_raw)
log_df, final_cash = backtest_strategy(df_indicators, stock_code)
# 🔥 使用新函数
sub_pdf = generate_sub_pdf_plus(df_indicators, log_df, stock_name, stock_code, final_cash, font_name)
sub_pdf_paths.append(sub_pdf)
stock_info_list.append((stock_name, stock_code))
if sub_pdf_paths:
merge_to_total_pdf(sub_pdf_paths, stock_info_list, font_name)
print("🎉 全部完成！含红♥历史信号 / 蓝♥未来节点 / 最优价 / PDF 嵌入")
# ================= 自动重训 & 胜率监控 =================
import tushare as ts
from sklearn.linear_model import LogisticRegression
from sklearn.metrics import roc_auc_score
import joblib, datetime as dt, numpy as np
ROLL_WINDOW = 240 # 滚动 240 日
RETRAIN_WEEKDAY = 4 # 周五
MIN_SIGNAL_COUNT = 20
WIN_THRESHOLD = 0.08
IC_DECAY_ALERT = 0.30
MONTE_PATHS = 5000
MONTE_CONF = 0.80
WEBHOOK_URL = "" # 飞书 webhook
def get_trade_days(start, end):
"""本地交易日历（零积分）1990-2030"""
holidays = [
'20241001','20241002','20241003','20241004','20241007',
'20250203','20250204','20250205','20250206','20250207',
'20250501','20250502','20250609','20250610',
'20251001','20251002','20251003','20251006','20251007',
'20260101','20260216','20260217','20260218','20260219','20260220'
]
bdays = pd.bdate_range(start, end)
hdays = pd.to_datetime(holidays)
return bdays.drop(bdays.intersection(hdays))
def calc_win_rate(df_raw, signal_col='buy_signal', ret_col='fwd_ret'):
"""计算胜率百分比"""
df = df_raw.copy()
df[ret_col] = df['close'].shift(-5) / df['close'] - 1
sig = df[signal_col].fillna(False)
if sig.sum() < MIN_SIGNAL_COUNT:
return None, None, None
win = (df[ret_col] > WIN_THRESHOLD) & sig
win_rate = win.sum() / sig.sum()
return win_rate, sig.sum(), df[ret_col][sig].mean()
def monte_carlo_nodes(last_price, mu, sigma, last_day, end_day, mask_days):
"""蒙特卡洛外推 + 可信度"""
days = len(mask_days)
dt1 = 1/252
paths = np.exp((mu - 0.5*sigma**2)*dt1 + sigma*np.sqrt(dt1)*np.random.normal(0,1,(MONTE_PATHS,days)))
prices = last_price * np.cumprod(paths,axis=1)
slopes = (prices[:,10:] - prices[:,:-10]) / prices[:,:-10] / 10 * 100
hit = ((slopes >= 0.5) & (slopes <= 2)) | (slopes > 2)
first_hit = np.full(MONTE_PATHS, np.nan)
for i in range(MONTE_PATHS):
hh = np.where(hit[i])[0]
if len(hh) > 0:
first_hit[i] = hh[0]
if np.all(np.isnan(first_hit)):
return pd.DataFrame()
p50 = np.nanpercentile(first_hit, 50)
p80 = np.nanpercentile(first_hit, (1-MONTE_CONF)*100)
date50 = mask_days[int(p50)]
date80 = mask_days[int(p80)]
price50 = last_price * (1 + mu)**int(p50)
price80 = last_price * (1 + mu)**int(p80)
return pd.DataFrame({'date':[date50, date80],
'price':[price50, price80],
'credib':[0.50, MONTE_CONF]})
def auto_retrain(df):
"""周五滚动重训逻辑回归"""
if pd.Timestamp.today().weekday() != RETRAIN_WEEKDAY:
return None
train = df.tail(ROLL_WINDOW).copy()
train['ret5'] = train['close'].shift(-5) / train['close'] - 1
train = train.dropna()
if len(train) < MIN_SIGNAL_COUNT or train['ret5'].gt(WIN_THRESHOLD).sum() < MIN_SIGNAL_COUNT//2:
return None
X = train[['slope10', 'macd', 'vol_ratio']].fillna(0)
y = (train['ret5'] > WIN_THRESHOLD).astype(int)
model = LogisticRegression().fit(X, y)
auc = roc_auc_score(y, model.predict_proba(X)[:, 1])
joblib.dump(model, 'signal_model.pkl')
print(f"[RETRAIN] {pd.Timestamp.today().date()} 完成 AUC={auc:.3f}")
return auc
def push_lark(msg):
if not WEBHOOK_URL:
return
import requests
requests.post(WEBHOOK_URL, json={"msg_type":"text","content":{"text":msg}})
# ------------- 胜率 & 蒙特卡洛可信度 -------------
win_rate, sig_cnt, avg_ret = calc_win_rate(df_indicators)
model_auc = auto_retrain(df_indicators)
mask_days = get_trade_days(df_indicators.index[-1].strftime('%Y%m%d'), '20251231')
mc_df = monte_carlo_nodes(
df_indicators['close'].iloc[-1],
df_indicators['close'].pct_change().mean() * 252,
df_indicators['close'].pct_change().std() * np.sqrt(252),
df_indicators.index[-1],
'20251231',
mask_days
)
elements.append(Spacer(1, 12))
elements.append(Paragraph("<b>5. 胜率与蒙特卡洛可信度</b>", styles['Heading2']))
if win_rate is None:
elements.append(Paragraph("• 信号样本不足，暂无法计算胜率。", styles['Normal']))
else:
elements.append(Paragraph(
f"• 历史胜率：<b>{win_rate:.1%}</b> 信号次数：<b>{sig_cnt}</b> 平均收益：<b>{avg_ret:.2%}</b>",
styles['Normal']))
if mc_df.empty:
elements.append(Paragraph("• 蒙特卡洛外推：未探测到可信度 ≥80% 节点。", styles['Normal']))
else:
mc_tbl_data = [["日期", "预估价(元)", "可信度"]] + \
[[d.strftime('%Y-%m-%d'), f"{p:.2f}", f"{c:.0%}"] for d, p, c in mc_df.values]
mc_tbl = Table(mc_tbl_data, colWidths=[120, 100, 80], repeatRows=1)
mc_tbl.setStyle(TableStyle([
("BACKGROUND", (0, 0), (-1, 0), colors.lightgreen),
("TEXTCOLOR", (0, 0), (-1, -1), colors.black),
("ALIGN", (0, 0), (-1, -1), "CENTER"),
("FONTNAME", (0, 0), (-1, -1), font_name),
("GRID", (0, 0), (-1, -1), 1, colors.black),
]))
elements.append(mc_tbl)
push_lark(f"{stock_name}({stock_code}) 胜率={(win_rate or 0):.1%} 信号={sig_cnt or 0} 可信度节点={len(mc_df)}")
import seaborn as sns
import matplotlib.pyplot as plt
plt.rcdefaults() # 先复位
sns.set_theme(
style='whitegrid', # 清爽网格
palette='Set2', # 柔和 8 色
font='Microsoft YaHei', # 系统雅黑
rc={'figure.figsize': (16, 9), 'figure.dpi': 150,
'axes.titlesize': 16, 'axes.labelsize': 14,
'xtick.labelsize': 11, 'ytick.labelsize': 11,
'legend.fontsize': 11, 'figure.subplot.left': 0.08,
'figure.subplot.right': 0.95, 'figure.subplot.bottom': 0.10,
'figure.subplot.top': 0.92}
)
# ---------------------------------------------------------
# 5. 统一入口
# ---------------------------------------------------------
if __name__ == "__main__":
main_plus()
需要优化吗，对生成的PDF进行美观度调整，看上去更加VIP主题效果更好。对数据的整合要直观易懂
扫一扫
2925
被折叠的 条评论
为什么被折叠?
到【灌水乐园】发言
