huanzhang12 / atla_robust_rl Goto Github PK

Hi, when I want to extract my best adversary model, why:
python: can't open file 'get_best_pickle.py': [Errno 2] No such file or directory

Hi, when i use the command "python run.py --config-path config_halfcheetah_sappo_convex.json"
It will use the auto_LiRPA package and I have installed this package.
It will have an error:
File "atla/src/policy_gradients/steps.py", line 796, in robust_ppo_step
stdev=stdev).mean()
File "atla/src/policy_gradients/convex_relaxation.py", line 100, in get_kl_bound
ilb, iub = model.compute_bounds(inputs, IBP=True, C=None, method=None, bound_lower=True, bound_upper=True)
File "anaconda3/lib/python3.7/site-packages/auto_LiRPA-0.1-py3.7.egg/auto_LiRPA/bound_general.py", line 592, in compute_bounds
lower, upper = self._IBP_general(node=final, C=C)
File "anaconda3/lib/python3.7/site-packages/auto_LiRPA-0.1-py3.7.egg/auto_LiRPA/bound_general.py", line 763, in _IBP_general
node.interval = node.interval_propagate(*inp, C=C)
File "anaconda3/lib/python3.7/site-packages/auto_LiRPA-0.1-py3.7.egg/auto_LiRPA/bound_ops.py", line 694, in interval_propagate
center, deviation = BoundLinear._propogate_Linf(h_L, h_U, w)
File "anaconda3/lib/python3.7/site-packages/auto_LiRPA-0.1-py3.7.egg/auto_LiRPA/bound_ops.py", line 647, in _propogate_Linf
center = torch.bmm(mid.unsqueeze(1), w.transpose(-1, -2)).squeeze(1)
RuntimeError: Expected tensor to have size 64 at dimension 0, but got size 1 for argument #2 'batch2' (while checking arguments for bmm)

Could you give some idea to fix this auto_LiRPA bug?