As "genetic programming for quant", gpquant is a modification of the genetic algorithm package gplearn in Python, used for factor mining.
Functions that calculate factors are implemented using the functional class Function
, which includes 23 basic functions and 37 time series functions. All functions are essentially scalar functions, but because vectorized computation is used, both inputs and outputs are in vector form.
Fitness evaluation indicators are implemented using the functional class Fitness
, which includes several fitness functions, mainly the Sharpe Ratio ("sharpe_ratio").
The vectorized factor backtesting framework follows the logic of first using the defined strategy function to turn the received "factor" into a "signal", and then using the signal processing function to turn the signal into an "asset" to implement backtesting. These two steps are combined in the functional class Backtester
.
The formula tree is used to write the calculation formula of the factor in prefix notation, and is represented using the formula tree SyntaxTree
. Each formula tree represents a factor, and is composed of Node
's; each Node
contains its own data, parent node, and child nodes. The Node
's own data can be a Function
, variable, constant, or time-series constant.
The formula tree can be crossed over subtree mutated, hoisted, point mutated or reproduced (logic can be referred to gplearn).
It contains the symbolic regression class (SymbolicRegressor
). gpquant
essentially uses genetic algorithms to solve the symbolic regression problem, and defines some parameters during the genetic process, such as population size and number of generations.
Download the gpquant package (pip install gpquant) and import the SymbolicRegressor class.
Like the example in gplearn
, performing symbolic regression on
import numpy as np
import pandas as pd
from matplotlib import pyplot as plt
from sklearn.utils import *
from gpquant.SymbolicRegressor import SymbolicRegressor
# Step 1
x0 = np.arange(-1, 1, 1 / 10.0)
x1 = np.arange(-1, 1, 1 / 10.0)
x0, x1 = np.meshgrid(x0, x1)
y_truth = x0**2 - x1**2 + x1 - 1
ax = plt.figure().gca(projection="3d")
ax.set_xlim(-1, 1)
ax.set_ylim(-1, 1)
surf = ax.plot_surface(x0, x1, y_truth, rstride=1, cstride=1, color="green", alpha=0.5)
plt.show()
# Step 2
rng = check_random_state(0)
# training samples
X_train = rng.uniform(-1, 1, 100).reshape(50, 2)
y_train = X_train[:, 0] ** 2 - X_train[:, 1] ** 2 + X_train[:, 1] - 1
X_train = pd.DataFrame(X_train, columns=["X0", "X1"])
y_train = pd.Series(y_train)
# testing samples
X_test = rng.uniform(-1, 1, 100).reshape(50, 2)
y_test = X_test[:, 0] ** 2 - X_test[:, 1] ** 2 + X_test[:, 1] - 1
# Step 3
sr = SymbolicRegressor(
population_size=2000,
tournament_size=20,
generations=20,
stopping_criteria=0.01,
p_crossover=0.7,
p_subtree_mutate=0.1,
p_hoist_mutate=0.1,
p_point_mutate=0.05,
init_depth=(6, 8),
init_method="half and half",
function_set=["add", "sub", "mul", "div", "square"],
variable_set=["X0", "X1"],
const_range=(0, 1),
ts_const_range=(0, 1),
build_preference=[0.75, 0.75],
metric="mean absolute error",
parsimony_coefficient=0.01,
)
sr.fit(X_train, y_train)
# Step 4
print(sr.best_estimator)
gpquant是对Python的遗传算法包gplearn的一个改造,用于进行因子挖掘
计算因子的函数,用仿函数类Function实现了23个基本函数和37个时间序列函数。所有的函数本质上都是标量函数,但因为采用了向量化计算,所以输入和输出都是向量形式
适应度评价指标,用仿函数类Fitness实现了几个适应度函数,主要是应用其中的夏普比率sharpe_ratio
向量化的因子回测框架,逻辑是先根据定义的策略函数把拿到的因子factor变成信号signal,再通过信号处理函数把信号signal变成资产asset实现回测,这两步统一在仿函数Backtester类里实现
公式树,把因子的计算公式写成前缀表达式,然后用公式树SyntaxTree表示。每一个公式树代表一个因子,由节点Node构成;每个Node存放了自身数据、父节点和子节点。节点的自身数据可以是Function、变量、常量,或者时间序列常数
公式树可以交叉crossover、子树突变subtree_mutate、提升突变hoist_mutate、点突变point_mutate或者繁殖reproduce(逻辑可参照gplearn)
符号回归类,gpquant因子挖掘本质上是用遗传算法解决符号回归问题,其中定义了遗传过程中的一些参数,如种群数量population_size、遗传代数generations等
下载gpquant包(pip install gpquant),导入SymbolicRegressor类
from gpquant.SymbolicRegressor import SymbolicRegressor
跟gplearn一样的例子,把$y=X_0^2 - X_1^2 + X_1 - 1$对$X_0$和$X_1$进行符号回归,大约在第9代能找到正确答案
import numpy as np
import pandas as pd
from matplotlib import pyplot as plt
from sklearn.utils import *
from gpquant.SymbolicRegressor import SymbolicRegressor
# Step 1
x0 = np.arange(-1, 1, 1 / 10.0)
x1 = np.arange(-1, 1, 1 / 10.0)
x0, x1 = np.meshgrid(x0, x1)
y_truth = x0**2 - x1**2 + x1 - 1
ax = plt.figure().gca(projection="3d")
ax.set_xlim(-1, 1)
ax.set_ylim(-1, 1)
surf = ax.plot_surface(x0, x1, y_truth, rstride=1, cstride=1, color="green", alpha=0.5)
plt.show()
# Step 2
rng = check_random_state(0)
# training samples
X_train = rng.uniform(-1, 1, 100).reshape(50, 2)
y_train = X_train[:, 0] ** 2 - X_train[:, 1] ** 2 + X_train[:, 1] - 1
X_train = pd.DataFrame(X_train, columns=["X0", "X1"])
y_train = pd.Series(y_train)
# testing samples
X_test = rng.uniform(-1, 1, 100).reshape(50, 2)
y_test = X_test[:, 0] ** 2 - X_test[:, 1] ** 2 + X_test[:, 1] - 1
# Step 3
sr = SymbolicRegressor(
population_size=2000,
tournament_size=20,
generations=20,
stopping_criteria=0.01,
p_crossover=0.7,
p_subtree_mutate=0.1,
p_hoist_mutate=0.1,
p_point_mutate=0.05,
init_depth=(6, 8),
init_method="half and half",
function_set=["add", "sub", "mul", "div", "square"],
variable_set=["X0", "X1"],
const_range=(0, 1),
ts_const_range=(0, 1),
build_preference=[0.75, 0.75],
metric="mean absolute error",
parsimony_coefficient=0.01,
)
sr.fit(X_train, y_train)
# Step 4
print(sr.best_estimator)