emmanuel-naive / matd3 Goto Github PK

I'm sorry to bother you, but I encountered the following problem while learning the code you created: 1. No matter how I modify the reward function, the agent's network updates and starts spinning in place; 2. The training process was originally normal, but after a certain number of times, it suddenly converged at a very low reward value, resulting in poor training effectiveness. I hope you can provide me with some ideas to solve the problem in your free time. Thank you

Hello, as a newcomer to reinforcement learning, I have some questions I would like to ask.

1. Are the hyperparameters in the model suitable for all scenarios?
2. When I was training the model with your parameters, I found that the final result was very poor, which appeared in almost every scenario. I recorded the reward value in each episode, and I found that the model does not converge to a high reward, on the contrary，it converges to a negative reward every time. So I have a doubt, are the hyperparameters in the code you provided final hyperparameters? If I need to train the model myself, in what aspects do I need to adjust the parameters?
   Thank you very much!