3 ms·
Unlike normal models with clear inputs and outputs with numerous methods of validating robustness of results, reinforcement learning requires a lot more dedicat
by minimaxir 8y ago
Unlike normal models with clear inputs and outputs with numerous methods of validating robustness of results, reinforcement learning requires a lot more dedication for a chance at the model converging and actually learning, with substantially more training time/cost needed.
Even with a higher-level library, it's not plug-and-play on new games (although tools like OpenAI Gym make it substantially easier), and you'll still need to do a lot of tuning.