Merge pull request #301 from WentseChen/crafter_release

update_crafter_hyper-para

examples/crafter/crafter_ppo.yaml

@@ -1,7 +1,10 @@
 seed: 0
 lr: 7e-4
 critic_lr: 7e-4
-episode_length: 256
+episode_length: 512
+ppo_epoch: 5
+data_chunk_length: 8
+num_mini_batch: 8
 run_dir: ./run_results/
 experiment_name: train_crafter
 log_interval: 10

openrl/envs/crafter/crafter.py

@@ -52,6 +52,8 @@ def __init__(
         )
 
         self.action_space = gym.spaces.Discrete(self.env.action_space.n)
+
+        self.rand_seed = 42
 
     def step(self, action: int):
         obs, reward, done, truncated, info = self.env.step(action)
@@ -65,7 +67,13 @@ def reset(
         options: Optional[Dict[str, Any]] = None,
         **kwargs
     ):
-        obs, info = self.env.reset(seed, options)
+
+        if seed is not None:
+            self.rand_seed = seed
+        else:
+            self.rand_seed += 1
+
+        obs, info = self.env.reset(self.rand_seed, options)
         obs = self.convert_observation(obs)
 
         return obs, info