diff --git a/mindspore_rl/algorithm/maddpg/maddpg.py b/mindspore_rl/algorithm/maddpg/maddpg.py
index 2ad91392..ba7e0377 100644
--- a/mindspore_rl/algorithm/maddpg/maddpg.py
+++ b/mindspore_rl/algorithm/maddpg/maddpg.py
@@ -240,8 +240,8 @@ def __init__(self, params):
         self.actor_net = params.get('actor_net')
 
         # optimizer network
-        critic_optimizer = nn.Adam(self.critic_net.trainable_params(), learning_rate=params.get('learning_rate'))
-        actor_optimizer = nn.Adam(self.actor_net.trainable_params(), learning_rate=params.get('learning_rate'))
+        critic_optimizer = nn.Adam(self.critic_net.trainable_params(), learning_rate=params.get('learning_rate'), eps=1e-5)
+        actor_optimizer = nn.Adam(self.actor_net.trainable_params(), learning_rate=params.get('learning_rate'), eps=1e-5)
 
         # loss network
         self.target_actor_net = params.get('target_actor_net')
diff --git a/mindspore_rl/algorithm/maddpg/maddpg_trainer.py b/mindspore_rl/algorithm/maddpg/maddpg_trainer.py
index 9948c80e..31da8908 100644
--- a/mindspore_rl/algorithm/maddpg/maddpg_trainer.py
+++ b/mindspore_rl/algorithm/maddpg/maddpg_trainer.py
@@ -96,7 +96,7 @@ def train(self, episodes, callbacks=None, ckpt_path=None):
                         and represent for `loss, rewards, steps, [optional]others.` in order"
                     )
                 episode_rewards.append(float(rewards.asnumpy()))
-                if i % 1000 == 0:
+                if (i % 1000 == 0) and (i != 0):
                     print("-----------------------------------------")
                     # pylint: disable=C0209
                     print(
diff --git a/mindspore_rl/algorithm/ppo/ppo.py b/mindspore_rl/algorithm/ppo/ppo.py
index 2495fa91..6246c0bc 100644
--- a/mindspore_rl/algorithm/ppo/ppo.py
+++ b/mindspore_rl/algorithm/ppo/ppo.py
@@ -273,7 +273,7 @@ def __init__(self, params):
         trainable_parameter = (
             self.critic_net.trainable_params() + self.actor_net.trainable_params()
         )
-        optimizer_ppo = nn.Adam(trainable_parameter, learning_rate=params["lr"])
+        optimizer_ppo = nn.Adam(trainable_parameter, learning_rate=params["lr"], eps=1e-5)
         ppo_loss_net = self.PPOLossCell(
             self.actor_net,
             self.critic_net,
diff --git a/mindspore_rl/algorithm/qmix/qmix.py b/mindspore_rl/algorithm/qmix/qmix.py
index 772990f4..7759be02 100644
--- a/mindspore_rl/algorithm/qmix/qmix.py
+++ b/mindspore_rl/algorithm/qmix/qmix.py
@@ -813,7 +813,7 @@ def __init__(self, params):
         trainable_params = (
             self.policy_net.trainable_params() + self.mixer_net.trainable_params()
         )
-        optimizer = nn.Adam(trainable_params, learning_rate=params["lr"])
+        optimizer = nn.Adam(trainable_params, learning_rate=params["lr"], eps=1e-5)
 
         qmix_loss_cell = self.QMIXLossCell(
             params, self.policy_net, self.mixer_net, self.target_mixer_net
diff --git a/mindspore_rl/environment/petting_zoo_mpe_environment.py b/mindspore_rl/environment/petting_zoo_mpe_environment.py
index 6c1c1695..09711f4c 100644
--- a/mindspore_rl/environment/petting_zoo_mpe_environment.py
+++ b/mindspore_rl/environment/petting_zoo_mpe_environment.py
@@ -87,7 +87,7 @@ def __init__(self, params, env_id=0):
         observation_space = gym2ms_adapter(list(self._env.observation_spaces.values()))
         env_action_space = self._env.action_spaces["agent_0"]
         action_space = Space(
-            (env_action_space.n,), env_action_space.dtype.type, batch_shape=(self._num,)
+            (env_action_space.n,), np.float32, batch_shape=(self._num,)
         )
         reward_space = Space((1,), np.float32, batch_shape=(self._num,))
         done_space = Space((1,), np.bool_, low=0, high=2, batch_shape=(self._num,))