mtenv 0.1

MTEnv: Environment interface for mulit-task reinforcement learning

MTEnv

Documentation

How to add new environments

There are two workflows:

• The user have a standard gym environment, which they want to convert into a multitask environment. E.g.: examples/bandit.py has a BanditEnv which is a standard multi-arm bandit, without any explicit notion of task. The user has the following options:

  • Write a new subclass, say MTBanditEnv (which subclasses MTEnv) as shown in examples/mtenv_bandit.py.

  • Use the MTEnvWrapper and wrap the existing standard model class. An example is shown in examples/wrapped_bandit.py.

• If the user does not have a standard gym environment to start with, it is recommended that they directly extend the MTEnv class.

Running examples

• pip install -r requirements.txt
• Alternatively, feel free to use my conda env: source activate source /private/home/sodhani/.conda/envs/mtenv/bin/activate /private/home/sodhani/.conda/envs/mtenv
• In the root folder, run PYTHONPATH=. python examples/<filename>.py.

Pending items

1. Google Doc to track some ongoing work.