vdblm / Human-Machine-MDP

Learning to Switch Between Machines and Humans

Requirements

To install requirements:

pip install -r requirements.txt

Code structure (Training and Evaluation)

• agents/

  • agents/hum_mac_agents.py contains human/machine action policies.

  • agents/switching_agents.py contains implementation of Algorithm 1, Algorithm 2, UCRL2, and the Greedy algorithm in the paper.

• environments/ contains the code to produce all the environment types (episodic MDPs) used in the paper (i.e., Env-1, Env-2, and Env-3).

• experiments/ contains all the known/unknown human experiments. For example,

  • SensorBasedSwitchingExperiment.run_unknown_human_exp trains and evaluates Algorithm 2, UCRL2, and the Greedy algorithm in an episodic setting.

• plot/ contains the code to plot the figures in the paper.

Results

See the notebook plots.ipynb to reproduce the results in the paper:

jupyter notebook plots.ipynb