Hands-On Data Science with Anaconda

Reinforcement learning

Reinforcement learning is a wonderful research area of machine learning. It has a root from behavioral psychology. The mechanism would maximize some notion of cumulative reward when certain actions were taken in a set of environments (that is, an agent tries to learn optimal behavior through trial-and-error interactions within a dynamic environment setting).

Let's use an R package called ReinforcementLearning. First, let's look at the dataset, shown here:

> library("ReinforcementLearning") 
> set.seed(123) 
> data <- sampleGridSequence(1000) 
> dim(data) 
[1] 1000    4 
> head(data) 
  State Action Reward NextState 
1    s2   left     -1        s2 
2    s4  right     -1        s4 
3    s2   down     -1        s2 
4    s4     up     -1        s4 
5    s4     up     -1        s4 
6    s1   left     -1        s1 > unique(data$State) ...