[機械学習・進化計算による株式取引最適化] No.07-04 実行と結果

このプログラムの目的

このプログラムはnsga2_neuroevo.pyを呼び出し，取引最適化を開始させることを目的とします．

また，手法・環境のパラメータやデータセットのパスなどをここで指定します．

work_share
├07_evolutionary_algorithms
  ├Dockerfile
  ├docker-compose.yml
  └src
    ├result(自動生成)
    ├neuro_evolution_torch
    | ├nsga2_neuroevo.py
    | ├torch_model.py
    | ├simulation_cython.pyx
    | └setup.py
    └experiment01.py (これを作成)

実験プログラム

今回の実験では次のようなパラメータを設定しました．

import pandas as pd
import os
import json

import neuro_evolution_torch.nsga2_neuroevo as neuroevo

import warnings
warnings.simplefilter('ignore')

def run(args):
    with open(f'{args["result_dir"]}/args.json', 'w') as f:
        json.dump(args, f, indent=4, ensure_ascii=False)

    learner = neuroevo.NSGA2_Learner(
        max_generation=args['max_generation'],
        seed=args['seed'],
        pop_size=args['pop_size'],
        init_money=args['init_money'],
        device=args['device'],
        num_select=args['num_select'],
        train_start_date=args['train_start_date'],
        train_end_date=args['train_end_date'],
        re_production_type=args['re_production_type'])

    learner.learning(args)

def experiment1():
    dataset_list = [
        ('../dataset_pred/non_fitting_0.3', 'nf0.3'),
        ('../dataset_pred/non_fitting_0.5', 'nf0.5'),
        ('../dataset_pred/non_fitting_1.0', 'nf1.0'),
    ]

    training_term_list = [
        ('2017-01-01', '2022-01-01')
    ]
    for train_start_date, train_end_date in training_term_list:
        for dataset_dir, tag1 in dataset_list:
            result_dir = f'result/experiment01_1/{train_start_date}_{train_end_date}_{tag1}'
            #result_dir = f'result/experiment09/test_{tag1}'
            for run_seed in range(1):
                result_seed_dir = f'{result_dir}/{run_seed}'
                os.makedirs(result_seed_dir, exist_ok=True)
                args = {
                    'result_dir':result_seed_dir,
                    'dataset_dir':dataset_dir,
                    'init_money':10_000_000,
                    'max_generation':200,
                    'pop_size':500,
                    'seed':run_seed,
                    'save_interval':100,
                    'device':'cuda',
                    'num_select':9,
                    'train_start_date':train_start_date,
                    'train_end_date':train_end_date,
                    're_production_type':'crossover'
                }
                run(args)

if __name__ == '__main__':
    experiment1()