note.py

import sys
import os

sys.path.insert(0, "/mnt/lustre/chenyun/jpx_trading/jpx_trading_byRL")
from data.utils import FeatureEngineer, data_split, load_dataset
import pickle
from sklearn import preprocessing
import numpy as np
import matplotlib.pyplot as plt
import pandas as pd
from matplotlib.pyplot import MultipleLocator
import random
import json

from gym.envs.registration import register
from algo.ppo import trainer
from config.trading_config import generate_config
import argparse
parser = argparse.ArgumentParser(description='manual to this script')
parser.add_argument("--seed", type=int, default=0)
parser.add_argument("--cuda", action="store_true")
args = parser.parse_args()
seed_num = args.seed
cuda = args.cuda
with open("/mnt/lustre/chenyun/jpx_trading/jpx_trading_byRL/data/data_scale.pkl", 'rb') as f:
    df = pickle.load(f)
df = df.sort_values(['date', 'tic']).reset_index(drop=True)

train = data_split(df, '2017-01-05', '2020-11-22')
test = data_split(df, '2020-11-22', '2021-12-03')

df.index = df['date'].factorize()[0]
whole_data = df

print(whole_data)

# Process your data here [doing data  cleaning, features engineering here]

feature_list = ['open', 'high', 'low', 'close', 'volume', 'macd', 'rsi_30', 'cci_30', 'dx_30']
feature_ws=[30, 30, 30, 30, 30, 1, 1, 1, 1]

stock_num = len(train.tic.unique())
stock_feature_dimension = len(feature_list)
print(f"Stock Feature Dimension: {stock_feature_dimension}, State Space: {(stock_num, sum(feature_ws))}")
feature_dimension = len(feature_list)
print(f"Feature Dimension: {feature_dimension}")
exp_name = 'trading_strategy_jpx_onppo_seed'+ str(seed_num)
env_train_kwargs={
    "exp_name": exp_name,
    "df":train,
    "feature_ws":feature_ws,
    "stock_num":stock_num,
    "tech_indicator_list":feature_list,
    "reward_scaling":1e-1
}
run_whole_train_kwargs={
    "exp_name": exp_name,
    "df":whole_data,
    "feature_ws":feature_ws,
    "stock_num":stock_num,
    "tech_indicator_list":feature_list,
    "reward_scaling":1e-1,
    "run_whole_train":True
}

main_config, create_config = generate_config(exp_name=exp_name)


register(id='trading-v2', entry_point='env.portfolio_env:StockPortfolioEnvJpx', max_episode_steps=10000)

if __name__ == '__main__':
    main_config.policy.cuda = cuda
    main_config.policy.model.obs_shape = [100, sum(main_config.feature_ws)]
    env_train_kwargs["feature_ws"] = main_config.feature_ws
    run_whole_train_kwargs["feature_ws"] = main_config.feature_ws
    print(json.dumps(main_config, indent=4))
    trainer('trading-v2', main_config, create_config, env_train_kwargs, run_whole_train_kwargs, seed=seed_num)