NELQlearning

nel subcommittee code

Getting it running

The recommended version of Python is 2.7

1. Install the nel-framework according to https://github.com/eaplatanios/nel_framework/#installation-instructions

2. Install the dependencies using pip install torch and for python2 pip install enum34

3. Run train.py to train, and test.py to visualize

Making a new agent

Create a new class that extends BaseAgent. For this agent, implement the following functions:

• save(): Saves the state of the agent. If you don't want to save, just use a pass statement for the body
• _load(): Loads the state of the agent. If you don't want to load, just use a pass statement for the body
• step(): You can add whatever parameters you want to this. This will be called in your training loop. This function should call env.step which takes in a reference to the agent and a function that takes in no arguments and returns the next move that your agent should do.
• next_move(): You can add whatever parameters you want to this. This function should return the next move that the agent should do.

Note: The recommended way you should structure the interaction between the environment and the agent is by creating a function that takes in the parameters needed for next_move() and returns a partial function that takes in no arguments that can be called by env.step. See the example in RLAgent.

Visualisation

Youtube video [paper]

Presentation

Class presentation [link]