tradeX 0.0.2

Machine learning based crypto currency price prediction

Intel-Trace

This project provides tool to download Binance trading data in the past. Use the data to develop a mini AI model for coin prediction. Follow below step to reproduce:

1. Install Python and required packages

pip install -r requirements.txt

Other lacking library can be installed manually by pip install <lib_name> command.

2. Download Binance Kline-Candlestick Data

cd tradeX
# ex1: install BTC data from Sep 08 2019 to now, interval is one hour
PYTHONPATH=. python scripts/get_all_time_data.py --symbol BTCUSDT --interval 1h

# ex2: install other coin pairs data, .e.g Solona
PYTHONPATH=. python scripts/get_all_time_data.py --symbol SOLUSDT --interval 3m

# ex3: install data from a specified time to now, if not passed, default value is 1567962000000
PYTHONPATH=. python scripts/get_all_time_data.py --symbol BTCUSDT --dfrom <time_in_milliseconds> --interval 1h

Downloaded data is saved in data directory as default. Try to play with other interval values (1m 3m 5m 15m 30m 1h 2h 4h 6h 8h 12h 1d 3d 1w 1M).

3. Split train-val data

cd tradeX

# replace the data path below with the newly downloaded data file.
PYTHONPATH=. python scripts/split_train_val.py --data data/BTCUSDT_3m_2022.10.23_14.24.54.json

After running above script, there will be 2 files ending with *_train.json and *_val.json generated.

4. Train a simple model

cd tradeX

# Without GPU
python scripts/train.py 
--train_data <data/*_train.json file path> 
--val_data <data/*_val.json file path>
--window_size <50 or 100 or any positive number, default 50>
--epochs 30
--bsize 8
  
# With GPU
CUDA_VISIBLE_DEVICES=0,1,2,3 PYTHONPATH=. python scripts/train.py 
--train_data <data/*_train.json file path> 
--val_data <data/*_val.json file path>
--window_size <50 or 100 or any positive number, default 50>
--epochs 30
--bsize 8
--num_gpus 3

Trained models is saved in weights directory with latest timestamp.

5. Eval model on validation set

cd tradeX

python scripts/eval.py 
--val_data data/BTCUSDT_3m_2022.10.23_14.24.54_val.json
--window_size <50 or any positive number, should be same with trained window_size>
--weight weights/2022.10.23_15.50.09/epoch=19-val_loss=3.94.pth.ckpt

6. Auto futures-trading on Binance with your own strategy

python run_testnet.py --interval 3m --symbol btcusdt