Open oowadanoko opened 5 months ago
Hi, since carracing is an image-input task, it is not appropriate to use MLP as an approximation function, we added a carracing-only training script ( dsac_cnn_carracing_offasync.py ) that uses CNN ( cnn.py ), and we also added a new A task variant of carracing, carracingraw (gym_carracingraw_data.py)
Thank you, it works now, thanks a lot!
It seems caused by the shape of observation_space.
obs_dim
is a tuple(4, 96, 96)
, butact_dim
is a scalar3
. I use default arguments (same with example_train/main.py). Should I use some different arguments for running this task? Your papar got an excellent result in this task.