My goal is to create an ML model to be able to classify different game stages, e.g., dialog with a non-player character, exploration, combat with enemy, in-game menu etc.

In order to do that, I am looking for an agent pre-trained on such a game. I am intending to develop a model using this pre-trained agent to produce a data set (frames-labels) and finally I will use that data set to train a model to classify those different stages.

I could only find a pre-trained model for the Doom, however, it is not much appropriate for my case because it does not have different game stages (it is merely based on running & shooting). Training my own Reinforcement Learning Agent is a whole another workload in terms of both time and GPU such a game needs.

Any single idea could help me a lot. Thanks!

  • 1
    $\begingroup$ Welcome to AI:SE! These 2 repos have a collection of pre-trained RL agents on a variety of games: github.com/DLR-RM/rl-baselines3-zoo and github.com/araffin/rl-baselines-zoo $\endgroup$ – mugoh Nov 8 '20 at 19:42
  • $\begingroup$ Hello @mugoh thank you for your answer! I cannot see any game related my criteria above. Instead of atari based or simple games, I need an agent trained on more complex game. $\endgroup$ – bbasaran Nov 8 '20 at 23:22
  • $\begingroup$ My thought is that, at a high level, the problems you identify (dialogue, exploration, combat) all have different structures, and would require different techniques. (Exploration, for instance, is pathfinding. Dialogue is natural language processing. Combat is more like a card game with hidden information.) So I doubt there is any single ML function that can accomplish them all, but it's an good question, imho, for that reason. $\endgroup$ – DukeZhou Nov 17 '20 at 3:39

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.