state_a = np.array(new_data, dtype=np.float).flatten()
state = torch.from_numpy(state_a).float().unsqueeze_(dim=0)
if torch.cuda.is_available() and self.net_spec["gpu"]:
state = state.cuda()
pdparam = self.calc_pdparam(state, evaluate=False)
// use multi-policy. note arg change
After Change
"""Non-atomizable act to override agent.act(), do a single pass on the entire state_a instead of composing body_act"""
// gather and flatten
states = []
for (e, b), body in util.ndenumerate_nonan(self.agent.body_a):
state = state_a[(e, b)]
states.append(state)
state = torch.tensor(states).view(-1).unsqueeze_(0).float()
if torch.cuda.is_available() and self.net.gpu:
state = state.cuda()
pdparam = self.calc_pdparam(state, evaluate=False)
