hotfix(nyz): fix lunarlander dqn config and get formatted config

ding/config/utils.py

@@ -349,12 +349,59 @@ def save_config_formatted(config_: dict, path: str = 'formatted_total_config.py'
                                 f.write("            replay_buffer=dict(\n")
                                 for k4, v4 in v3.items():
                                     if (k4 != 'monitor' and k4 != 'thruput_controller'):
-                                        if (isinstance(v4, str)):
-                                            f.write("                {}='{}',\n".format(k4, v4))
-                                        elif v4 == float('inf'):
-                                            f.write("                {}=float('{}'),\n".format(k4, v4))
+                                        if (isinstance(v4, dict)):
+                                            f.write("                {}=dict(\n".format(k4))
+                                            for k5, v5 in v4.items():
+                                                if (isinstance(v5, str)):
+                                                    f.write("                    {}='{}',\n".format(k5, v5))
+                                                elif v5 == float('inf'):
+                                                    f.write("                    {}=float('{}'),\n".format(k5, v5))
+                                                elif (isinstance(v5, dict)):
+                                                    f.write("                    {}=dict(\n".format(k5))
+                                                    for k6, v6 in v5.items():
+                                                        if (isinstance(v6, str)):
+                                                            f.write("                        {}='{}',\n".format(k6, v6))
+                                                        elif v6 == float('inf'):
+                                                            f.write(
+                                                                "                        {}=float('{}'),\n".format(
+                                                                    k6, v6
+                                                                )
+                                                            )
+                                                        elif (isinstance(v6, dict)):
+                                                            f.write("                        {}=dict(\n".format(k6))
+                                                            for k7, v7 in v6.items():
+                                                                if (isinstance(v7, str)):
+                                                                    f.write(
+                                                                        "                            {}='{}',\n".format(
+                                                                            k7, v7
+                                                                        )
+                                                                    )
+                                                                elif v7 == float('inf'):
+                                                                    f.write(
+                                                                        "                            {}=float('{}'),\n".
+                                                                        format(k7, v7)
+                                                                    )
+                                                                else:
+                                                                    f.write(
+                                                                        "                            {}={},\n".format(
+                                                                            k7, v7
+                                                                        )
+                                                                    )
+                                                            f.write("                        ),\n")
+                                                        else:
+                                                            f.write("                        {}={},\n".format(k6, v6))
+                                                    f.write("                    ),\n")
+                                                else:
+                                                    f.write("                    {}={},\n".format(k5, v5))
+                                            f.write("                ),\n")
                                         else:
-                                            f.write("                {}={},\n".format(k4, v4))
+                                            if (isinstance(v4, str)):
+                                                f.write("                {}='{}',\n".format(k4, v4))
+                                            elif v4 == float('inf'):
+                                                f.write("                {}=float('{}'),\n".format(k4, v4))
+
+                                            else:
+                                                f.write("                {}={},\n".format(k4, v4))
                                     else:
                                         if (k4 == 'monitor'):
                                             f.write("                monitor=dict(\n")

dizoo/box2d/lunarlander/config/lunarlander_dqn_config.py

@@ -17,11 +17,10 @@ lunarlander_dqn_default_config = dict(
         cuda=False,
         # Whether the RL algorithm is on-policy or off-policy.
         on_policy=False,
-        # Model config used for model creating. Remember to change this, especially "obs_dim" and "action_dim" according to specific env.
         model=dict(
-            obs_dim=8,
-            action_dim=4,
-            encoder_hidden_dim_list=[512, 64],
+            obs_shape=8,
+            action_shape=4,
+            encoder_hidden_size_list=[512, 64],
             # Whether to use dueling head.
             dueling=True,
         ),
@@ -31,8 +30,6 @@ lunarlander_dqn_default_config = dict(
         nstep=nstep,
         # learn_mode config
         learn=dict(
-            # How many steps to train after collector's one collection. Bigger "train_iteration" means bigger off-policy.
-            # collect data -> train fixed steps -> collect data -> ...
             update_per_collect=10,
             batch_size=64,
             learning_rate=0.001,
@@ -55,7 +52,7 @@ lunarlander_dqn_default_config = dict(
                 type='exp',
                 start=0.95,
                 end=0.1,
-                decay=50_000,
+                decay=50000,
             ),
             replay_buffer=dict(replay_buffer_size=100000, )
         ),