OpenAI Gym is a open-source Python toolkit for developing and comparing reinforcement learning algorithms. This R package is a wrapper for the OpenAI Gym API, and enables access to an ever-growing variety of environments.

Installation

You can install the latest development version from CRAN:

install.packages("gym")

Or from GitHub with:

if (packageVersion("devtools") < 1.6) {
  install.packages("devtools")
}
devtools::install_github("paulhendricks/gym-R", subdir = "R")

If you encounter a clear bug, please file a minimal reproducible example on GitHub.

Getting started

Setting up the server

To download the code and install the requirements, you can run the following shell commands:

git clone https://github.com/openai/gym-http-api
cd gym-http-api
pip install -r requirements.txt

This code is intended to be run locally by a single user. The server runs in python.

To start the server from the command line, run this:

python gym_http_server.py

For more details, please see here: https://github.com/openai/gym-http-api.

Running an example in R

In a separate R terminal, you can then try running the example agent and see what happens:

library(gym)

remote_base <- "http://127.0.0.1:5000"
client <- create_GymClient(remote_base)
print(client)

# Create environment
env_id <- "CartPole-v0"
instance_id <- env_create(client, env_id)
print(instance_id)

# List all environments
all_envs <- env_list_all(client)
print(all_envs)

# Set up agent
action_space_info <- env_action_space_info(client, instance_id)
print(action_space_info)
agent <- random_discrete_agent(action_space_info[["n"]])

# Run experiment, with monitor
outdir <- "/tmp/random-agent-results"
env_monitor_start(client, instance_id, outdir, force = TRUE, resume = FALSE)

episode_count <- 100
max_steps <- 200
reward <- 0
done <- FALSE

for (i in 1:episode_count) {
  ob <- env_reset(client, instance_id)
  for (i in 1:max_steps) {
    action <- env_action_space_sample(client, instance_id)
    results <- env_step(client, instance_id, action, render = TRUE)
    if (results[["done"]]) break
  }
}

# Dump result info to disk
env_monitor_close(client, instance_id)

Citation

To cite package ‘gym’ in publications use:

Paul Hendricks (2016). gym: Provides Access to the OpenAI Gym API. R package version 0.1.0. https://CRAN.R-project.org/package=gym

A BibTeX entry for LaTeX users is

@Manual{,
  title = {gym: Provides Access to the OpenAI Gym API},
  author = {Paul Hendricks},
  year = {2016},
  note = {R package version 0.1.0},
  url = {https://CRAN.R-project.org/package=gym},
}