evgps / a3c_trading

Trading with recurrent actor-critic reinforcement learning

A3C trading

Note: Sorry for misleading naming - please use A3C_trading.py for training and test_trading.py for testing.

Trading with recurrent actor-critic reinforcement learning - check paper and more detailed old report

Configuration: config.py

This file contains all the pathes and gloabal variables to be set up

Dataset: download from GDrive

After setting config.py please run this file to download and preprocess the data need for training and evaluation

Environment: trader_gym.py

OpenAI.gym-like environment class

Model: A3C_class.py

This file is containing AC_network, Worker and Test_Worker classes

Training: A3C_training.py

Run this file, preferrable in tmux. During training it will create files in tensorboard_dir and in model_dir

Testing: A3C_testing.ipynb

Jupyter notebook contains all for picturing

Cite as:

@article{ponomarev2019using, title={Using Reinforcement Learning in the Algorithmic Trading Problem}, author={Ponomarev, ES and Oseledets, IV and Cichocki, AS}, journal={Journal of Communications Technology and Electronics}, volume={64}, number={12}, pages={1450--1457}, year={2019}, publisher={Springer} }

a3c_trading's Issues

potential forward looking

Thanks for sharing the great work!

1. In test_trading.py file - line 41: R = (R - R.mean()) / (R.max() - R.min())
   This normalization used the whole test set's data, will this introduce forward-looking bias? Since in living trading, we do not know the populations' mean/max/min (we can obtain rolling or historical mean/min/max).

2. Didn't find ['RTS-12.15']w5k, ['RTS-3.16', 'RTS-6.16']w5k these two files in G-drive. It would be great if you can share these two files and original data file

Thanks!

Результат не соответствуют заявленному.

Оформлено как научная работа, а код то не работает как описано.
С текущими настройками, ничего не меняя в коде, ревард сотни эпизодов в 10 потоках топчется на месте.
Зачем вводить людей в заблуждение, выкладывая такое ?

training output nan

fixed the issue. thansk

Missing A3C_training.py

Where can I find A3C_training.py file as said in README file. Thanks

Some vars not defined in config.py

In A3C_training.py

from configs import TRAIN_DATA, LOAD_MODEL, LR, FRAMES_STACKED, NUM_WORKERS, MODEL_DIR

But I can't find the TRAIN_DATA, FRAMES_STACKED and NUM_WORKERS in configs.py