I'm studying reinforcement learning in order to implement a kind of time series pattern analyzer such as market.

The most examples I have seen are based on the maze environment.

But in real market environment, the signal changes endlessly as time passes and I can not guess how can I model environment and states.

Another question is about buy-sell modeling.

Let's assume that the agent randomly buy at time $t$ and sell at time $t + \alpha$.

It's simple to calculate reward. The problem is how can I model $Q$ matrix and how can I model signals between buy and sell actions.

Can you share some source code or guidance for similar situation?

  • $\begingroup$ This is a quite open-ended question. Can you be more specific about what problem you are considering, how RL applies, what you have done so far and what your specific confusion is? $\endgroup$ – Sean Owen Aug 21 '14 at 11:53
  • $\begingroup$ Please ask just one question per post, unless the questions are closely related. Your questions are quite different. $\endgroup$ – MrMeritology Aug 22 '14 at 15:25