Recently, I have heard of the concept of the reinforcement learning and I have got interested in it. So I decided to start a project that uses this kind of machine learning for constructing algorithms from existing knowledge. My problem is that I really don't know all the possible states and actions in order to predefine them earlier. Is there any way of avoiding this and some kind of automatic algorithms that explore all of them? Thank you for the patience.
$\begingroup$
$\endgroup$
Add a comment
|
1 Answer
$\begingroup$
$\endgroup$
When the set of states is not given or known it is used Fitted Value Iteration. http://cs229.stanford.edu/notes/cs229-notes12.pdf