Recommender_system_via_deep_RL icon indicating copy to clipboard operation
Recommender_system_via_deep_RL copied to clipboard

The problem of model training

Open nayujia opened this issue 3 years ago • 1 comments

Dear author, I am reproducing this code recently. I have some questions to ask you:

  1. Why is the model training "MAX_EPISODE_NUM" set to 8000 ? Is the result better the more times you train? Will it overfit?
  2. Have you set up offline experiments and online experiments?

nayujia avatar Nov 22 '21 10:11 nayujia

Dear author, I am reproducing this code recently. I have some questions to ask you:

  1. Why is the model training "MAX_EPISODE_NUM" set to 8000 ? Is the result better the more times you train? Will it overfit?
  2. Have you set up offline experiments and online experiments?

Hello, nayujia!

  1. There is no reason in 8000. You can change it. Even if you change the MAX_EPISODE_NUM to more than 8000, the result will not be getting better. I think the reason is in the simulator. check this reply https://github.com/backgom2357/Recommender_system_via_deep_RL/issues/5#issuecomment-963669449
  2. I just set up offline experiments.

backgom2357 avatar Nov 22 '21 14:11 backgom2357