Disable exploration during test phase

main.py

@@ -250,7 +250,7 @@ def update_belief_and_act(args, env, planner, transition_model, encoder, belief,
       belief, posterior_state, action = torch.zeros(args.test_episodes, args.belief_size, device=args.device), torch.zeros(args.test_episodes, args.state_size, device=args.device), torch.zeros(args.test_episodes, env.action_size, device=args.device)
       pbar = tqdm(range(args.max_episode_length // args.action_repeat))
       for t in pbar:
-        belief, posterior_state, action, next_observation, reward, done = update_belief_and_act(args, test_envs, planner, transition_model, encoder, belief, posterior_state, action, observation.to(device=args.device), test=False)
+        belief, posterior_state, action, next_observation, reward, done = update_belief_and_act(args, test_envs, planner, transition_model, encoder, belief, posterior_state, action, observation.to(device=args.device), test=True)
         total_rewards += reward.numpy()
         if not args.symbolic_env:  # Collect real vs. predicted frames for video
           video_frames.append(make_grid(torch.cat([observation, observation_model(belief, posterior_state).cpu()], dim=3) + 0.5, nrow=5).numpy())  # Decentre