thu-ml · Trinkle23897 · Aug 29, 2021 · Aug 29, 2021 · Aug 29, 2021 · Aug 29, 2021
diff --git a/.github/workflows/pytest.yml b/.github/workflows/pytest.yml
@@ -8,7 +8,7 @@ jobs:
     if: "!contains(github.event.head_commit.message, 'ci skip')"
     strategy:
       matrix:
-        python-version: [3.6, 3.7, 3.8]
+        python-version: [3.6, 3.7, 3.8, 3.9]
     steps:
     - uses: actions/checkout@v2
     - name: Set up Python ${{ matrix.python-version }}

diff --git a/docs/requirements.txt b/docs/requirements.txt
@@ -1,7 +1,8 @@
 gym
-tqdm
-torch
 numba
-tensorboard
+numpy>=1.20
 sphinx<4
 sphinxcontrib-bibtex
+tensorboard
+torch
+tqdm
diff --git a/test/continuous/test_ppo.py b/test/continuous/test_ppo.py
@@ -81,12 +81,12 @@ def test_ppo(args=get_args()):
         args.state_shape, hidden_sizes=args.hidden_sizes, device=args.device
     ), device=args.device).to(args.device)
     # orthogonal initialization
-    for m in list(actor.modules()) + list(critic.modules()):
+    for m in set(actor.modules()).union(critic.modules()):
         if isinstance(m, torch.nn.Linear):
             torch.nn.init.orthogonal_(m.weight)
             torch.nn.init.zeros_(m.bias)
     optim = torch.optim.Adam(
-        list(actor.parameters()) + list(critic.parameters()), lr=args.lr)
+        set(actor.parameters()).union(critic.parameters()), lr=args.lr)
 
     # replace DiagGuassian with Independent(Normal) which is equivalent
     # pass *logits to be consistent with policy.forward

diff --git a/test/discrete/test_a2c_with_il.py b/test/discrete/test_a2c_with_il.py
@@ -75,7 +75,7 @@ def test_a2c_with_il(args=get_args()):
     actor = Actor(net, args.action_shape, device=args.device).to(args.device)
     critic = Critic(net, device=args.device).to(args.device)
     optim = torch.optim.Adam(
-        list(actor.parameters()) + list(critic.parameters()), lr=args.lr)
+        set(actor.parameters()).union(critic.parameters()), lr=args.lr)
     dist = torch.distributions.Categorical
     policy = A2CPolicy(
         actor, critic, optim, dist,

diff --git a/test/discrete/test_ppo.py b/test/discrete/test_ppo.py
@@ -72,12 +72,12 @@ def test_ppo(args=get_args()):
     actor = Actor(net, args.action_shape, device=args.device).to(args.device)
     critic = Critic(net, device=args.device).to(args.device)
     # orthogonal initialization
-    for m in list(actor.modules()) + list(critic.modules()):
+    for m in set(actor.modules()).union(critic.modules()):
         if isinstance(m, torch.nn.Linear):
             torch.nn.init.orthogonal_(m.weight)
             torch.nn.init.zeros_(m.bias)
     optim = torch.optim.Adam(
-        list(actor.parameters()) + list(critic.parameters()), lr=args.lr)
+        set(actor.parameters()).union(critic.parameters()), lr=args.lr)
     dist = torch.distributions.Categorical
     policy = PPOPolicy(
         actor, critic, optim, dist,

diff --git a/test/discrete/test_qrdqn.py b/test/discrete/test_qrdqn.py
@@ -18,7 +18,7 @@
 def get_args():
     parser = argparse.ArgumentParser()
     parser.add_argument('--task', type=str, default='CartPole-v0')
-    parser.add_argument('--seed', type=int, default=0)
+    parser.add_argument('--seed', type=int, default=1)
     parser.add_argument('--eps-test', type=float, default=0.05)
     parser.add_argument('--eps-train', type=float, default=0.1)
     parser.add_argument('--buffer-size', type=int, default=20000)

diff --git a/tianshou/policy/modelfree/a2c.py b/tianshou/policy/modelfree/a2c.py
@@ -129,7 +129,7 @@ def learn(  # type: ignore
                 loss.backward()
                 if self._grad_norm:  # clip large gradient
                     nn.utils.clip_grad_norm_(
-                        list(self.actor.parameters()) + list(self.critic.parameters()),
+                        set(self.actor.parameters()).union(self.critic.parameters()),
                         max_norm=self._grad_norm)
                 self.optim.step()
                 actor_losses.append(actor_loss.item())

diff --git a/tianshou/policy/modelfree/ppo.py b/tianshou/policy/modelfree/ppo.py
@@ -139,7 +139,7 @@ def learn(  # type: ignore
                 loss.backward()
                 if self._grad_norm:  # clip large gradient
                     nn.utils.clip_grad_norm_(
-                        list(self.actor.parameters()) + list(self.critic.parameters()),
+                        set(self.actor.parameters()).union(self.critic.parameters()),
                         max_norm=self._grad_norm)
                 self.optim.step()
                 clip_losses.append(clip_loss.item())