Trade-Bot-2

This project tests a continuous stock trading environment for reinforcement learning. Use TradingEnv9.

• Base parameters:
  • buffer_size = 1,000,000; batch_size = 100; gamma = 0.99; tau = 0.00001; policy_freq = 2; lr = 0.001; policy_noise = 0.2; noise_clip = 0.5; expl_noise = 0.15; starting_step = 20,000;
• TD3_TradingEnv9_main_42 parameters:
  • Initialization: . ; init_thresh = 2.0; starting_step = 30,000;
  • Robust: . ; lr = 0.0001; reg = 0.01; starting_step = 15,000;
• TD3_TradingEnv9_main_86 parameters:
  • Initialization: . ; init_thresh = 2.0;
  • Robust: . ; lr = 0.0005; reg = 0.001; starting_step = 10,000;
• TD3_TradingEnv9_main_70 parameters:
  • Initialization: . ; init_thresh = 5.0;
  • Robust: . ; reg = 0.0001; starting_step = 10,000;

Changes:

• train__tts.ipynb implements a new process for training and testing in which training is done on short periods of stock history sampled from a distribution skewed toward the start date of testing
• TradingEnvNorm Introduces normalization into the tts training process
  • Creates a new problem of "data drift"