DQN_based_BTC_Trading_Algorithm

background

Most modern trading algorithms are dominated by CNNs and LSTMs.

LSTM is suitable for processing time series data. However, as CNNs became more adept at identifying patterns in charts, the two algorithms became dominant. This algorithm aims to utilize the advantages of both models by processing time series using the POMDP definition of DQN and identifying patterns using CNN.

Setup

Python 3.8 (for dataclass support) or higher is required.

pip install -r requirements.txt

or

pip install h5py
pip install keras

Dataset

download from HERE

Result

baseline

Ours