mirror of
https://github.com/shy3130/tick-stock-panel.git
synced 2026-11-02 10:26:13 +00:00
shy3130
43dfef3e62
买卖穿越成交共用 rolling_mean(close,5) 作参考线, 该线含当日收盘 — 盘中模拟成交价依赖 15:00 才确定的收盘价, minute_fill 声称的分钟级 精确成交被系统性污染; 卖出信号回放侧 (build_minute_exit_reference) 早已用代数法剔除当根, 双侧纪律不一致。 新增 build_minute_entry_reference: (window·MA - close)/(window-1) 代数剔除当根 = 前 window-1 根均值, 当日开盘即已知; 与卖出侧同住 minute_trigger 模块, 「当日已知参考线」单一实现。strategy.py 三处 调用点统一替换, 无重复公式。回归测试钉住性质: 同一盘中路径只改 当日收盘, 成交价必须不变。
69 lines
2.8 KiB
Python
69 lines
2.8 KiB
Python
"""分钟级成交回放的参考价计算 — 买卖两侧同守「当日已知」纪律。
|
|
|
|
参考线在当日开盘即必须已知: 含当根收盘的均线会让盘中模拟成交价依赖
|
|
15:00 才确定的收盘价 (前视), 污染 minute_fill 声称的分钟级精确成交。
|
|
"""
|
|
from __future__ import annotations
|
|
|
|
import numpy as np
|
|
|
|
MINUTE_EXIT_TRIGGER_SIGNALS = frozenset({
|
|
"signal_ma5_breakdown",
|
|
"signal_ma10_breakdown",
|
|
"signal_ma20_breakdown",
|
|
"signal_ma_dead_5_20",
|
|
})
|
|
|
|
|
|
def unsupported_minute_exit_signals(signals: list[str] | tuple[str, ...]) -> list[str]:
|
|
return sorted(set(signals) - MINUTE_EXIT_TRIGGER_SIGNALS)
|
|
|
|
|
|
def build_minute_exit_reference(
|
|
close: np.ndarray,
|
|
fields: dict[str, np.ndarray],
|
|
exit_signal_code: np.ndarray,
|
|
exit_signal_ids: tuple[str, ...],
|
|
) -> np.ndarray:
|
|
"""为可回放的卖出信号计算当日已知的价格触发线。"""
|
|
result = np.full(close.shape, np.nan, dtype=np.float32)
|
|
|
|
def _apply(code: int, value: np.ndarray) -> None:
|
|
mask = (exit_signal_code == code) & np.isfinite(value) & (value > 0)
|
|
result[mask] = value[mask].astype(np.float32)
|
|
|
|
with np.errstate(divide="ignore", invalid="ignore"):
|
|
for code, signal_id in enumerate(exit_signal_ids):
|
|
if signal_id == "signal_ma5_breakdown" and "ma5" in fields:
|
|
_apply(code, (5.0 * fields["ma5"] - close) / 4.0)
|
|
elif signal_id == "signal_ma10_breakdown" and "ma10" in fields:
|
|
_apply(code, (10.0 * fields["ma10"] - close) / 9.0)
|
|
elif signal_id == "signal_ma20_breakdown" and "ma20" in fields:
|
|
_apply(code, (20.0 * fields["ma20"] - close) / 19.0)
|
|
elif (
|
|
signal_id == "signal_ma_dead_5_20"
|
|
and "ma5" in fields
|
|
and "ma20" in fields
|
|
):
|
|
sum4 = 5.0 * fields["ma5"] - close
|
|
sum19 = 20.0 * fields["ma20"] - close
|
|
_apply(code, (sum19 - 4.0 * sum4) / 3.0)
|
|
|
|
result.setflags(write=False)
|
|
return result
|
|
|
|
|
|
def build_minute_entry_reference(close: np.ndarray, window: int = 5) -> np.ndarray:
|
|
"""分钟穿越成交的参考线: 前 window-1 根收盘均值 (当日开盘即已知)。
|
|
|
|
rolling_mean 第 t 行含当根收盘, 代数剔除当根即得前 window-1 根均值:
|
|
(window·MA_window[t] - close[t]) / (window-1)。与卖出侧
|
|
build_minute_exit_reference 同一纪律 — 买卖参考线都不得使用当日收盘。
|
|
MA 不足窗口 (NaN) 的行结果为 NaN, 调用方按无参考线退化到 VWAP。
|
|
"""
|
|
from app.backtest.matrix import rolling_mean # 延迟导入: matrix 模块级依赖本模块
|
|
|
|
ma = rolling_mean(close, window)
|
|
with np.errstate(divide="ignore", invalid="ignore"):
|
|
return (float(window) * ma - close) / float(window - 1)
|