ZM-Learn / L2RPN_WCCI_a_Solution

Repository for L2RPN WCCI 2020 competition. One possible solution.
23 stars 8 forks source link

How many episodes does it take to train ? #2

Open ShanxiWuYanZu opened 1 year ago

ShanxiWuYanZu commented 1 year ago

Hello author,I noticed that your code has 3000 training episodes,But I found during my training that 200 episodes seemed to convergence (or less),So I want to ask if it really needs 3000 episodes of training. After all, it takes a long time to compute and I have limited computing resources,Looking forward to your reply.

ZM-Learn commented 1 year ago

Hello, Sorry for the late reply, my open-source repositories are out of maintenance for a long time. If I remember correctly, the datasets for the experiments are much larger than the local datasets. More episodes are needed when sampling training datasets from a wider range. If you find the model converges for your local machine and local datasets, you can stop training and use the model for testing.

ShanxiWuYanZu commented 1 year ago

Thanks for your reply. I found that 200 episodes seemed to converge only on the training set, but the test effect was not good, so I will increase the training amount. Thanks again for your reply