I am solving a real-world problem to make self adaptive decisions while using context.I am using reinforcement learning to address this problem but formulating a reward function is a big challenge for me. For instance:i need to consider human environment (context) and have to make decision based on user activities in building.Does the condition here reflects the continuous or discrete state/action space?
I take account of statistical function(ex Bayesian,
Markov) or count method generally. This statistical function or count method provide efficient factor.
This function should réflex for episodes the way in which the process has a success.
Hello,
Here I let you an interesting page that gives some advises on how to define a proper reward function:
https://es.mathworks.com/help/reinforcement-learning/ug/define-reward-signals.html?s_tid=mwa_osa_a
Muhammad Ali Normally, reward functions are probabilistic and can be implemented using Markov Chains to remember state transitions. I recommend using matlab as it has built in libraries for reinforcement learning techniques.
- Badges
- Science topic
- Similar topics
- Philosophy
- Philosophy Of Science
- Reasoning