Documentation

rlR: (Deep) Reinforcement learning in R

Installation

R package installation

devtools::install_github("smilesun/rlR")

or

devtools::install_github("smilesun/rlR", dependencies = TRUE)

Python dependency

rlR use keras with tensorflow as its backend for neural network as functional approximator and OpenAI gym.

see Python Dependencies Installation and Configuration

Example of Neural Network as Functional Approximator

Choose an environment to learn

library(rlR)
## - type 'reticulate::py_discover_config()' to check default python
## - to use a different python path, execute the following after package is loaded:
## reticulate::use_python('/path/to/your/python')
env = makeGymEnv("CartPole-v0")
env
## 
## action cnt: 2 
## state original dim: 4 
## discrete action

If you have R package “imager” installed, you could get a snapshot of the environment by

env$snapshot(preprocess = F)

Initialize agent with the environment

agent = initAgent("AgentDQN", env)
agent$learn(200L)

Look at the performance

agent$plotPerf(F)

Specify a task to be sovled by creating your own Environment

see Custom Environment