Experiment File
- class or_suite.experiment.experiment.Experiment(env, agent, dict)[source]
Optional instrumentation for running an experiment.
Runs a simulation between an arbitrary openAI Gym environment and an algorithm, saving a dataset of (reward, time, space) complexity across each episode, and optionally saves trajectory information.
- seed
random seed set to allow reproducibility
- dirPath
(string) location to store the data files
- nEps
(int) number of episodes for the simulation
- deBug
(bool) boolean, when set to true causes the algorithm to print information to the command line
- env
(openAI env) the environment to run the simulations on
- epLen
(int) the length of each episode
- numIters
(int) the number of iterations of (nEps, epLen) pairs to iterate over with the environment
- save_trajectory
(bool) boolean, when set to true saves the entire trajectory information
- render_flag
(bool) boolean, when set to true renders the simulations
- agent
(or_suite.agent.Agent) an algorithm to run the experiments with
- data
(np.array) an array saving the metrics along the sample paths (rewards, time, space)
- trajectory_data
(list) a list saving the trajectory information