Series of n-armed bandit environments for the OpenAI Gym
This code is inspired by Jesse Cooper's work: https://github.com/JKCooper2/gym-bandits
The environments added in this repository are based on Wang et. al experiments described in the paper Learning to Reinforcement Learn. https://arxiv.org/abs/1611.05763#
Each environment uses a different set of:
- Probability Distributions - A list of probabilities of the likelihood that a particular bandit will pay out
- Reward Distributions - A list of either rewards (if number) or means and standard deviations (if list) of the payout that bandit has
E.g. BanditTwoArmedHighLowFixed-v0 has p_dist=[0.8, 0.2], r_dist=[1, 1], meaning 80% of the time that action 0 is
selected it will payout 1, and 20% of the time action 2 is selected it will payout 1
You can access the distributions through the p_dist and r_dist variables using env.p_dist or env.r_dist if you want to match
your weights against the true values for plotting results of various algorithms
To fit the universe-starter-agent, the observation of the bandits has been modified from 0 (type: gym.spaces.Discrete) to [0] (type: gym.spaces.box.Box).
Some of the environments return pieces of information regarding the arms. For example: the index of the optimal arm or the value of a parameter.
New in this repository:
- BanditTwoArmedIndependentUniform-v0: The two arms return a reward of 1 with probabilities p1 and p2 ~ U[0,1]
- BanditTwoArmedDependentUniform-v0: The first arm returns a reward of 1 with probability p ~ U[0,1], the second arm with probability 1-p
- BanditTwoArmedDependentEasy-v0: The first arm returns a reward of 1 with probability p ~ U{0.1,0.9}, the second arm with probability 1-p
- BanditTwoArmedDependentMedium-v0: The first arm returns a reward of 1 with probability p ~ U{0.25,0.75}, the second arm with probability 1-p
- BanditTwoArmedDependentHard-v0: The first arm returns a reward of 1 with probability p ~ U{0.4,0.6}, the second arm with probability 1-p
- BanditElevenArmedWithIndex: One optimal arm always returns a reward of 5, the other arms a reward of 1.1 ; The 11th arm return a reward of 0.1*
Other environments:
- BanditTwoArmedDeterministicFixed-v0: Simplest case where one bandit always pays, and the other always doesn't
- BanditTwoArmedHighLowFixed-v0: Stochastic version with a large difference between which bandit pays out of two choices
- BanditTwoArmedHighHighFixed-v0: Stochastic version with a small difference between which bandit pays where both are good
- BanditTwoArmedLowLowFixed-v0: Stochastic version with a small difference between which bandit pays where both are bad
- BanditTenArmedRandomFixed-v0: 10 armed bandit with random probabilities assigned to payouts
- BanditTenArmedRandomRandom-v0: 10 armed bandit with random probabilities assigned to both payouts and rewards
- BanditTenArmedUniformDistributedReward-v0: 10 armed bandit with that always pays out with a reward selected from a uniform distribution
- BanditTenArmedGaussian-v0: 10 armed bandit mentioned on page 30 of Reinforcement Learning: An Introduction (Sutton and Barto)
git clone https://github.com/ThomasLecat/gym-bandit-environments.git
cd gym-bandits
pip install -e .
To install using requirements.txt or environment.yml call:
git+https://github.com/contimatteo/gym-multi-armed-bandit#egg=gym-bandits
In your gym environment
import gym_bandits
env = gym.make("BanditTenArmedGaussian-v0") # Replace with relevant env