reinforcement learning - Validation worse than training

Question

In OpenAI gym classic-like env training, the model yields good results and completes the task. Validating with unseen data yields considerably lower results. Tried:

• Added (+/- 2% random range) noise to observations to prevent the model from memorizing (both training/validation).
• Applied same normalization (fit to training, transform to training/validation), and also no normalization.

No matter, getting same results with the above. Any idea what it could be or what I can try?

question from:https://stackoverflow.com/questions/66060521/validation-worse-than-training

Answer 1

Waitting for answers