factor minner
This commit is contained in:
297
factor_mining/operators.py
Normal file
297
factor_mining/operators.py
Normal file
@@ -0,0 +1,297 @@
|
||||
"""
|
||||
算子系统：基础数学算子和技术指标算子的注册与管理
|
||||
支持算子的注册、查询、反射调用
|
||||
"""
|
||||
|
||||
import numpy as np
|
||||
import pandas as pd
|
||||
from typing import Dict, Callable, List, Optional, Any
|
||||
from abc import ABC, abstractmethod
|
||||
import inspect
|
||||
|
||||
|
||||
class Operator(ABC):
|
||||
"""算子基类"""
|
||||
|
||||
def __init__(self, name: str, func: Callable, description: str = ""):
|
||||
"""
|
||||
Parameters:
|
||||
-----------
|
||||
name : str
|
||||
算子名称（唯一标识）
|
||||
func : Callable
|
||||
算子函数
|
||||
description : str
|
||||
算子描述
|
||||
"""
|
||||
self.name = name
|
||||
self.func = func
|
||||
self.description = description
|
||||
self._signature = inspect.signature(func)
|
||||
|
||||
def __call__(self, *args, **kwargs):
|
||||
"""调用算子函数"""
|
||||
return self.func(*args, **kwargs)
|
||||
|
||||
def get_signature(self):
|
||||
"""获取函数签名"""
|
||||
return self._signature
|
||||
|
||||
def __repr__(self):
|
||||
return f"Operator(name='{self.name}', description='{self.description}')"
|
||||
|
||||
|
||||
class OperatorRegistry:
|
||||
"""算子注册表"""
|
||||
|
||||
def __init__(self):
|
||||
self._operators: Dict[str, Operator] = {}
|
||||
|
||||
def register(self, operator: Operator):
|
||||
"""注册算子"""
|
||||
if operator.name in self._operators:
|
||||
raise ValueError(f"算子 '{operator.name}' 已存在")
|
||||
self._operators[operator.name] = operator
|
||||
|
||||
def register_function(self, name: str, func: Callable, description: str = ""):
|
||||
"""直接注册函数为算子"""
|
||||
operator = Operator(name, func, description)
|
||||
self.register(operator)
|
||||
|
||||
def get(self, name: str) -> Optional[Operator]:
|
||||
"""获取算子"""
|
||||
return self._operators.get(name)
|
||||
|
||||
def has(self, name: str) -> bool:
|
||||
"""检查算子是否存在"""
|
||||
return name in self._operators
|
||||
|
||||
def list_all(self) -> List[str]:
|
||||
"""列出所有算子名称"""
|
||||
return list(self._operators.keys())
|
||||
|
||||
def get_all(self) -> Dict[str, Operator]:
|
||||
"""获取所有算子"""
|
||||
return self._operators.copy()
|
||||
|
||||
|
||||
# 全局算子注册表
|
||||
_registry = OperatorRegistry()
|
||||
|
||||
|
||||
def register_operator(name: str, description: str = ""):
|
||||
"""装饰器：注册算子"""
|
||||
|
||||
def decorator(func: Callable):
|
||||
_registry.register_function(name, func, description)
|
||||
return func
|
||||
|
||||
return decorator
|
||||
|
||||
|
||||
def get_operator(name: str) -> Optional[Operator]:
|
||||
"""获取算子"""
|
||||
return _registry.get(name)
|
||||
|
||||
|
||||
def get_registry() -> OperatorRegistry:
|
||||
"""获取全局注册表"""
|
||||
return _registry
|
||||
|
||||
|
||||
# ==================== 基础数学算子 ====================
|
||||
|
||||
|
||||
@register_operator("add", "加法: x + y")
|
||||
def _add(x: np.ndarray, y: np.ndarray) -> np.ndarray:
|
||||
return x + y
|
||||
|
||||
|
||||
@register_operator("sub", "减法: x - y")
|
||||
def _sub(x: np.ndarray, y: np.ndarray) -> np.ndarray:
|
||||
return x - y
|
||||
|
||||
|
||||
@register_operator("mul", "乘法: x * y")
|
||||
def _mul(x: np.ndarray, y: np.ndarray) -> np.ndarray:
|
||||
return x * y
|
||||
|
||||
|
||||
@register_operator("div", "除法: x / y (安全除法)")
|
||||
def _div(x: np.ndarray, y: np.ndarray) -> np.ndarray:
|
||||
denom = np.where(np.abs(y) < 1e-12, np.nan, y)
|
||||
return x / denom
|
||||
|
||||
|
||||
@register_operator("neg", "取负: -x")
|
||||
def _neg(x: np.ndarray) -> np.ndarray:
|
||||
return np.negative(x)
|
||||
|
||||
|
||||
@register_operator("abs", "绝对值: |x|")
|
||||
def _abs(x: np.ndarray) -> np.ndarray:
|
||||
return np.abs(x)
|
||||
|
||||
|
||||
@register_operator("log", "对数: log(|x|)")
|
||||
def _log(x: np.ndarray) -> np.ndarray:
|
||||
return np.log(np.clip(np.abs(x), 1e-12, None))
|
||||
|
||||
|
||||
@register_operator("sqrt", "平方根: sqrt(x)")
|
||||
def _sqrt(x: np.ndarray) -> np.ndarray:
|
||||
return np.sqrt(np.clip(x, 0.0, None))
|
||||
|
||||
|
||||
@register_operator("pow", "幂运算: x^y (限制范围)")
|
||||
def _pow(x: np.ndarray, y: np.ndarray) -> np.ndarray:
|
||||
y_clip = np.clip(y, -3.0, 3.0)
|
||||
with np.errstate(over="ignore", invalid="ignore"):
|
||||
out = np.power(np.clip(x, -1e6, 1e6), y_clip)
|
||||
out[~np.isfinite(out)] = np.nan
|
||||
return out
|
||||
|
||||
|
||||
# ==================== 时间序列算子 ====================
|
||||
|
||||
|
||||
def _rolling_mean(x: np.ndarray, window: int) -> np.ndarray:
|
||||
s = pd.Series(x)
|
||||
return s.rolling(window, min_periods=max(2, window // 2)).mean().to_numpy()
|
||||
|
||||
|
||||
def _rolling_std(x: np.ndarray, window: int) -> np.ndarray:
|
||||
s = pd.Series(x)
|
||||
return s.rolling(window, min_periods=max(2, window // 2)).std().to_numpy()
|
||||
|
||||
|
||||
def _ts_delta(x: np.ndarray, period: int) -> np.ndarray:
|
||||
s = pd.Series(x)
|
||||
return s.diff(period).to_numpy()
|
||||
|
||||
|
||||
def _ts_rank(x: np.ndarray, window: int) -> np.ndarray:
|
||||
s = pd.Series(x)
|
||||
return (
|
||||
s.rolling(window, min_periods=max(2, window // 2))
|
||||
.apply(lambda a: pd.Series(a).rank(pct=True).iloc[-1], raw=False)
|
||||
.to_numpy()
|
||||
)
|
||||
|
||||
|
||||
def _delay(x: np.ndarray, period: int) -> np.ndarray:
|
||||
s = pd.Series(x)
|
||||
return s.shift(period).to_numpy()
|
||||
|
||||
|
||||
# 注册时间序列算子（带不同窗口）
|
||||
for w in (3, 6, 12, 24, 48, 96):
|
||||
_registry.register_function(
|
||||
f"sma{w}", lambda x, w=w: _rolling_mean(x, w), f"简单移动平均: SMA(x, {w})"
|
||||
)
|
||||
_registry.register_function(
|
||||
f"std{w}", lambda x, w=w: _rolling_std(x, w), f"滚动标准差: STD(x, {w})"
|
||||
)
|
||||
_registry.register_function(
|
||||
f"rank{w}", lambda x, w=w: _ts_rank(x, w), f"滚动排名: RANK(x, {w})"
|
||||
)
|
||||
_registry.register_function(
|
||||
f"delta{w}", lambda x, w=w: _ts_delta(x, w), f"差分: DELTA(x, {w})"
|
||||
)
|
||||
_registry.register_function(
|
||||
f"delay{w}", lambda x, w=w: _delay(x, w), f"延迟: DELAY(x, {w})"
|
||||
)
|
||||
|
||||
|
||||
# ==================== 因子公式解析与计算 ====================
|
||||
|
||||
|
||||
class FactorFormula:
|
||||
"""因子公式：支持序列化和反序列化"""
|
||||
|
||||
def __init__(self, expression: str, feature_names: List[str]):
|
||||
"""
|
||||
Parameters:
|
||||
-----------
|
||||
expression : str
|
||||
因子表达式（使用算子名称）
|
||||
feature_names : List[str]
|
||||
特征名称列表
|
||||
"""
|
||||
self.expression = expression
|
||||
self.feature_names = feature_names
|
||||
|
||||
def compute(self, features: Dict[str, np.ndarray]) -> np.ndarray:
|
||||
"""
|
||||
计算因子值
|
||||
|
||||
Parameters:
|
||||
-----------
|
||||
features : Dict[str, np.ndarray]
|
||||
特征字典，key为特征名称
|
||||
|
||||
Returns:
|
||||
--------
|
||||
np.ndarray: 因子值
|
||||
"""
|
||||
# 构建计算环境
|
||||
env = {}
|
||||
|
||||
# 添加特征
|
||||
for name in self.feature_names:
|
||||
if name not in features:
|
||||
raise KeyError(f"特征 '{name}' 不存在")
|
||||
env[name] = features[name]
|
||||
|
||||
# 添加算子
|
||||
for op_name in _registry.list_all():
|
||||
op = _registry.get(op_name)
|
||||
if op:
|
||||
env[op_name] = op.func
|
||||
|
||||
# 添加numpy和pandas（用于某些表达式）
|
||||
env["np"] = np
|
||||
env["pd"] = pd
|
||||
|
||||
# 执行表达式
|
||||
try:
|
||||
# 限制可用的内置函数
|
||||
safe_builtins = {
|
||||
"abs": abs,
|
||||
"min": min,
|
||||
"max": max,
|
||||
"sum": sum,
|
||||
"len": len,
|
||||
}
|
||||
result = eval(self.expression, {"__builtins__": safe_builtins}, env)
|
||||
|
||||
# 确保结果是numpy数组
|
||||
if not isinstance(result, np.ndarray):
|
||||
if isinstance(result, (int, float)):
|
||||
# 标量转换为数组（广播）
|
||||
result = np.full(len(features[self.feature_names[0]]), result)
|
||||
else:
|
||||
result = np.array(result)
|
||||
|
||||
# 确保长度一致
|
||||
expected_len = len(features[self.feature_names[0]])
|
||||
if len(result) != expected_len:
|
||||
raise ValueError(
|
||||
f"表达式结果长度 {len(result)} 与特征长度 {expected_len} 不匹配"
|
||||
)
|
||||
|
||||
return result
|
||||
except Exception as e:
|
||||
raise RuntimeError(f"计算因子表达式失败: {e}\n表达式: {self.expression}")
|
||||
|
||||
def to_dict(self) -> Dict:
|
||||
"""序列化为字典"""
|
||||
return {"expression": self.expression, "feature_names": self.feature_names}
|
||||
|
||||
@classmethod
|
||||
def from_dict(cls, data: Dict) -> "FactorFormula":
|
||||
"""从字典反序列化"""
|
||||
return cls(data["expression"], data["feature_names"])
|
||||
|
||||
def __repr__(self):
|
||||
return f"FactorFormula(expression='{self.expression}', features={self.feature_names})"
|
||||
Reference in New Issue
Block a user