This is the code for the environment benchmarks used in "Learning Environment Simulators from Sparse Signals" (Shavit 2017).
To use the environments defined in the package, navigate to the root of this project's directory and call:
python setup.py install
From now on, you can
import gym_mnist and then use gym's
gym.make() to construct the MNIST-game envs.
To interactively play the different environments, call
python test-env.py [environment_name]
For a list of the environment names, simply call
python test-env.py -h