rickstaa/stable-learning-control

Speed difference LAC minghoa and Machine_learning_control

Closed this issue · 1 comments

The LAC implementation of Minghoa seems to be a lot faster (see this issue. I noticed that this is caused by the censor board logging that is in the spinning up implementation. When I add logging of all the weights in the network for example the training becomes impossible slow. We should therefore only enable Tensorboard logging for development. Or we should log values less frequently for example only when the model is saved. We can use the Cprofiler to check where the seed bottlenecks are.

This was fixed in d571487