HKUST MAFM AI_Finance course project.

• Rainbow DQN
• Strategy Demo
• PPO Tricks
• Stable Baseline 3
• Stabel Basiline 3 Instruction

Model:

• ALSTM: attention + LSTM
• padding-maske + CNN + Attention + Dense-net

4 * 5 : [BTC, ETH, LTC, XRP] * [0 H L C V]

Testing Data set: 20160 data; 14 day; 2023-04-19--2023-05-02; 0 error days Training Data set: 838079 data; 582 day; 2021-09-14--2023-04-18; 1 error days

1. Whole model, 4 channel bitcoin, output 4 channel signal;
2. 4 model: each predict 1 bitcoin signal
3. Volume model: predict future volume

State:

• 6 ohlcvt, postion-value, cash-value, total-value, average-price?

Action:

• 1: 20action (-10, 10)
• 2: position

Rewards

• SharpRatio + Punishment of trading Volume
• Risk aversion function

Model:

• DQN
• PPO
• DDPG

• Check the data form
• Dataset Generator
  • Factors
  • Labels
• ALSTM Training
• RL Environment
• PPO
• DDPG