Vectorize your env

Gym vector

You still want your agent to perform better ?

Then, I suggest to use Vectorized Environment to parallelize several environments. It garantees having multiple simultaneous sources of data during the training.

import gymnasium as gym
import gym_trading_env

if __name__ == "__main__":
  envs = gym.vector.make(
    "MultiDatasetTradingEnv",
    dataset_dir = "preprocessed_data",
    num_envs = 3)
  print(envs.reset())

(array([[4.1329634e-04, 9.9952787e-01, 1.0001770e+00, 9.9935079e-01,
      2.4035616e-02, 0.0000000e+00],
     [4.2866479e-04, 9.9957150e-01, 1.0002638e+00, 9.9926108e-01,
      1.8846054e-01, 0.0000000e+00],
     [7.6002884e-04, 9.9924058e-01, 1.0001662e+00, 9.9836242e-01,
      7.7975184e-02, 0.0000000e+00]], dtype=float32), {'step': array([0, 0, 0]), '_step': array([ True,  True,  True]), 'date': array([numpy.datetime64('2023-01-07T23:00:00.000000000'),
     numpy.datetime64('2023-01-07T23:00:00.000000000'),
     numpy.datetime64('2023-01-07T23:00:00.000000000')], dtype=object), '_date': array([ True,  True,  True]), 'position_index': array([0, 0, 0]), '_position_index': array([ True,  True,  True]), 'position': array([0, 0, 0]), '_position': array([ True,  True,  True]), 'data_open': array([16936.  , 16936.31,  1263.11]), '_data_open': array([ True,  True,  True]), 'data_volume': array([2.88336096e+00, 3.60376360e+03, 5.95961140e+03]), '_data_volume': array([ True,  True,  True]), 'data_date_close': array([Timestamp('2023-01-08 00:00:00'), Timestamp('2023-01-08 00:00:00'),
     Timestamp('2023-01-08 00:00:00')], dtype=object), '_data_date_close': array([ True,  True,  True]), 'data_low': array([16933.  , 16931.05,  1262.  ]), '_data_low': array([ True,  True,  True]), 'data_close': array([16944.  , 16943.57,  1264.07]), '_data_close': array([ True,  True,  True]), 'data_high': array([16947.  , 16948.04,  1264.28]), '_data_high': array([ True,  True,  True]), 'portfolio_valuation': array([1000., 1000., 1000.]), '_portfolio_valuation': array([ True,  True,  True]), 'portfolio_distribution_asset': array([0, 0, 0]), '_portfolio_distribution_asset': array([ True,  True,  True]), 'portfolio_distribution_fiat': array([1000., 1000., 1000.]), '_portfolio_distribution_fiat': array([ True,  True,  True]), 'portfolio_distribution_borrowed_asset': array([0, 0, 0]), '_portfolio_distribution_borrowed_asset': array([ True,  True,  True]), 'portfolio_distribution_borrowed_fiat': array([0, 0, 0]), '_portfolio_distribution_borrowed_fiat': array([ True,  True,  True]), 'portfolio_distribution_interest_asset': array([0, 0, 0]), '_portfolio_distribution_interest_asset': array([ True,  True,  True]), 'portfolio_distribution_interest_fiat': array([0, 0, 0]), '_portfolio_distribution_interest_fiat': array([ True,  True,  True]), 'reward': array([0, 0, 0]), '_reward': array([ True,  True,  True])})

Note

Using if __name__ == "__main__": is recommended as you might encounter errors if you omit it.

Special cases

In some cases (Jupyter Notebooks), you might need to use the SyncVectorEnv object from gym to avoid crashes :

import gymnasium as gym
import gym_trading_env

def make_env():
  env = gym.make(
      "MultiDatasetTradingEnv",
      dataset_dir= "preprocessed_data",
  )
  return env
envs = gym.vector.SyncVectorEnv([lambda: make_env() for _ in range(3)])

Run the environments

observation, info = env.reset()
while True:
  actions = [0, 0, 0] # 3D List as we have 3 simultaneous environments
  observation, reward, done, truncated, info = env.step(actions)