多进程并行量化回测框架

天山老妖

已于 2024-10-15 08:30:40 修改

阅读量338

点赞数 1

CC 4.0 BY-SA版权

分类专栏：量化IT Python 文章标签：量化交易量化回测多进程并行

于 2024-10-15 08:28:23 首次发布

本文链接：https://blog.csdn.net/A642960662/article/details/142933741

量化IT 同时被 2 个专栏收录

40 篇文章

订阅专栏

Python

20 篇文章

订阅专栏

在这里插入图片描述

import optuna
from multiprocessing import  Process, Queue, Manager
import numpy as np
    
def backtest_func(queue:Queue):
    import random
    # 回测结果
    result = dict()
    result['annual_return_rate'] = random.random()
    # 可以增加其它策略业绩评价指标
    
    queue.put(result) # 回测结果增加到队列
    return result
        
def run_backtest_parallel(backtest_func, n_jobs):
    queue = Manager().Queue()
    # 多进程并行回测
    job_list = []
    for i in range(n_jobs):  
        job = Process(target=backtest_func, args=(queue,)) 
        job_list.append(job)
        job.start()
    for job in job_list:
        job.join()
    # 回测结果接收
    post_results = list()
    while (not queue.empty()):
        post_result = queue.get()
        post_results.append(post_result)
    return post_results


if __name__ == '__main__':
    # 执行多进程并行
    post_results = run_backtest_parallel(backtest_func=backtest_func, n_jobs=5)
    # 查看结果
    for post_result in post_results:
        print(post_result)

在这里插入图片描述

import optuna
from multiprocessing import  Process, Queue, Manager
import numpy as np
    
def backtest_func(trading_params:dict):
    import random
    period = trading_params['period']
    # 回测结果
    result = dict()
    result['annual_return_rate'] = random.random()
    # 可以增加其它策略业绩评价指标
    return result

def objective_func(trial, backtest_func, trading_params:dict, queue:Queue):
    # 交易参数更新
    trading_params_dict = dict()
    for key, value in trading_params.items():
        if key == 'period':
            low = value[0]
            high = value[-1]
            step = value[1] - value[0]
            trading_params_dict[key] = trial.suggest_int(name=key, low=low, high=high, step=step)
        elif key == 'limited_turning_rate':
            low = value[0]
            high = value[-1]
            step = round(value[1] - value[0], 2)
            trading_params_dict[key] = trial.suggest_float(name=key, low=low, high=high, step=step)
        else:
            trading_params_dict[key] = value

    result = backtest_func(trading_params_dict)
    queue.put(result)
    return result.get('annual_return_rate', 0.0) # 优化年化收益率

def optimize(backtest_func, n_trials, trading_parmas:dict, queue:Queue):
    study_name = 'optuna_optimize'
    storage_name = "sqlite:///{}.db".format(study_name)
    study = optuna.load_study(study_name=study_name, storage=storage_name)
    study.optimize(lambda trial: objective_func(trial, backtest_func=backtest_func, trading_params=trading_parmas, queue=queue), 
                   n_trials=n_trials, gc_after_trial=True)
        
def run_backtest_parallel(backtest_func, trading_params, n_trials, n_jobs):
    # 创建学习过程
    study_name = 'optuna_optimize'
    storage_name = "sqlite:///{}.db".format(study_name)
    study = optuna.create_study(study_name=study_name, direction='maximize', storage=storage_name, load_if_exists=True)
        
    queue = Manager().Queue()
    # 多进程并行回测
    job_list = []
    for i in range(n_jobs):  
        job = Process(target=optimize, args=(backtest_func, n_trials, trading_params, queue)) 
        job_list.append(job)
        job.start()
    for job in job_list:
        job.join()
    # 回测结果接收
    post_results = list()
    while (not queue.empty()):
        post_result = queue.get()
        post_results.append(post_result)
    return post_results


if __name__ == '__main__':
    # 交易参数设置
    trading_params = dict()
    PERIOD_LIST = list(range(1, 11, 1))   # 换仓周期
    TURNOVER_RATE_LIST = np.arange(0.3, 0.8, 0.01) # 换手率
    trading_params['period'] = PERIOD_LIST
    trading_params['limited_turning_rate'] = TURNOVER_RATE_LIST
    # 执行多进程并行
    post_results = run_backtest_parallel(backtest_func=backtest_func, trading_params=trading_params, n_trials=10, n_jobs=5)
    # 查看结果
    for post_result in post_results:
        print(post_result)