israelg99 / kindred

Versatile agents who cut like a samurai and sting like a butterfly.

Kindred

To adapt and win different environments, the agents ought to be versatile.

*The bold models solved the environment while the italic ones did not.

Cart Pole

• (dq_dnn.py) Double Q-learning with a Deep Q-Network using Boltzmann Q-Policy with a large Experience Replay.

Frozen Lake

• (dq_dnn.py) Double Q-learning with a Deep Q-Network using Decaying Epsilon Q-Policy with a large Experience Replay.
• (q_nn.py) Q-learning with a neural network using Epsilon Q-Policy.
• (q_table.py) Q-learning with a table using Epsilon Q-Policy (other policies available in the code).

Mountain Car

• (dq_dnn.py) Double Q-learning with a Deep Q-Network using Decaying Epsilon Q-Policy with a large Experience Replay.