Model-Free Reinforcement Learning

Performs model-free reinforcement learning in R. This implementation enables the learning of an optimal policy based on sample sequences consisting of states, actions and rewards. In addition, it supplies multiple predefined reinforcement learning algorithms, such as experience replay. Methodological details can be found in Sutton and Barto (1998) .

Reinforcement Learning

ReinforcementLearning performs model-free reinforcement learning in R. This implementation enables the learning of an optimal policy based on sample sequences consisting of states, actions and rewards. In addition, it supplies multiple predefined reinforcement learning algorithms, such as experience replay.

Overview

The most important functions of ReinforcementLearning are:

• Learning an optimal policy from a fixed set of a priori known transition samples
• Predefined learning rules and action selection modes
• A highly customizable framework for model-free reinforcement learning tasks

Installation

You can easily install the latest version of ReinforcementLearning with

install.packages("ReinforcementLearning")
 
# Alternatively, install the development version from GitHub:
# install.packages("devtools")
devtools::install_github("nproellochs/ReinforcementLearning")

Usage

This section shows the basic functionality of how to perform reinforcement learning. First, load the corresponding package ReinforcementLearning.

library(ReinforcementLearning)

The following example shows how to learn a reinforcement learning agent using input data in the form of sample sequences consisting of states, actions and rewards. The result of the learning process is a state-action table and an optimal policy that defines the best possible action in each state.

# Generate sample experience in the form of state transition tuples
data <- sampleGridSequence(N = 1000)
head(data)
#>   State Action Reward NextState
#> 1    s4   left     -1        s4
#> 2    s2  right     -1        s3
#> 3    s2  right     -1        s3
#> 4    s3   left     -1        s2
#> 5    s4     up     -1        s4
#> 6    s1   down     -1        s2
 
# Define reinforcement learning parameters
control <- list(alpha = 0.1, gamma = 0.1, epsilon = 0.1)
 
# Perform reinforcement learning
model <- ReinforcementLearning(data, s = "State", a = "Action", r = "Reward", 
                               s_new = "NextState", control = control)
 
# Print result
print(model)
#> State-Action function Q
#>          right        up        down       left
#> s1 -1.09619438 -1.098533 -1.00183072 -1.0978962
#> s2 -0.01980279 -1.097758 -1.00252228 -1.0037977
#> s3 -0.02335524  9.884394 -0.01722548 -0.9985081
#> s4 -1.09616040 -1.106392 -1.10548631 -1.1059655
#> 
#> Policy
#>      s1      s2      s3      s4 
#>  "down" "right"    "up" "right" 
#> 
#> Reward (last iteration)
#> [1] -263

License

ReinforcementLearning is released under the MIT License

Copyright (c) 2018 Nicolas Pröllochs & Stefan Feuerriegel