Is it possible apply Reinfocrement Learning to classify data?

I am currently working on a project with Deep and Reinforcement Learning, that deals with the classification of simple datasets such as Fisheriris, Wine quality, traffic signs, etc. I was already familiarized with DL, so I have had no issues using its toolbox. In the case of RL, I am knew and I have basic knowledge, but as far as I understood RL is generally used for control applications, so I have not found yet a way to perform this task, and I want to know if it is possible and if it is worty to try it?

Answer 1

[Emmanouil Tzorakoleftherakis]

If you already have a labeled dataset, supervised learning is the way to go. Reinforcement learning is more for cases where data is being generated on the fly from a simulated model/real system. RL can be applied in controls problems, as well as more general decision making problems such resource allocation, calibration, etc.

[Juan Carlos Rendón]

@Emmanouil Tzorakoleftherakis and if I had an unlabeled dataset? Would it be a better choice to use RL? Or just unsupervised learning?

[Emmanouil Tzorakoleftherakis]

how would you define the reward signal then? you would need some groundtruth if you want to do classification with rl

[Rebecca Nekou Malihi]

I actually have the same problem. I have labeled data, but the classes are simply too many to be traind with deep learning. there are 1441 diffrent classes. but i have 100 samples. also its more like controlling two degrees of freedom with 2000x8 epochs. so its more like to sovle the problem.so can you say if its possible to load data in a costume environment?