fixed env seeding in test_sac_with_il.py (thu-ml#1081)

Author: maxhuettenrauch

test/continuous/test_sac_with_il.py

@@ -78,6 +78,8 @@ def test_sac_with_il(args: argparse.Namespace = get_args()) -> None:
     # seed
     np.random.seed(args.seed)
     torch.manual_seed(args.seed)
+    train_envs.seed(args.seed)
+    test_envs.seed(args.seed + args.training_num)
     # model
     net = Net(args.state_shape, hidden_sizes=args.hidden_sizes, device=args.device)
     actor = ActorProb(net, args.action_shape, device=args.device, unbounded=True).to(args.device)
@@ -181,10 +183,12 @@ def stop_fn(mean_rewards: float) -> bool:
         action_scaling=True,
         action_bound_method="clip",
     )
+    il_test_env = gym.make(args.task)
+    il_test_env.reset(seed=args.seed + args.training_num + args.test_num)
     il_test_collector = Collector(
         il_policy,
         # envpool.make_gymnasium(args.task, num_envs=args.test_num, seed=args.seed),
-        gym.make(args.task),
+        il_test_env,
     )
     train_collector.reset()
     result = OffpolicyTrainer(