Closed gimme1dollar closed 1 year ago
Hi! You are right that it is not necessary for training. But having a test dataset helps to calculate test/model_loss (this is done in the evaluation step), which can be compared to train/model_loss to see if the world model overfits.
I got it. Thanks for your reply!
Hi! Thanks for your work on pytorch implementation of dreamerv2. It helps me a lot.
I am just curious on the reason for building two different datasets for the both train and test environments. It seems not crucial for the training, but why do we need a dataset in test env? Please let me know if this is obviously necessary and I am missing something.