diff --git a/examples/A2C/atari_agent.py b/examples/A2C/atari_agent.py
index 5604f71016538650b0ed0355dd6cd2856f52c60e..94d2125214a9ef52273763cea6cc0213cc34c963 100755
--- a/examples/A2C/atari_agent.py
+++ b/examples/A2C/atari_agent.py
@@ -71,7 +71,10 @@ class AtariAgent(parl.Agent):
             lr = layers.data(
                 name='lr', shape=[1], dtype='float32', append_batch_size=False)
             entropy_coeff = layers.data(
-                name='entropy_coeff', shape=[], dtype='float32')
+                name='entropy_coeff',
+                shape=[1],
+                dtype='float32',
+                append_batch_size=False)
 
             total_loss, pi_loss, vf_loss, entropy = self.alg.learn(
                 obs, actions, advantages, target_values, lr, entropy_coeff)
diff --git a/examples/IMPALA/atari_agent.py b/examples/IMPALA/atari_agent.py
index 98d4a4c4fd3ea611f60f2d8da850265025541b4b..0746f951f6920a70b0af87430af51879b635ada7 100755
--- a/examples/IMPALA/atari_agent.py
+++ b/examples/IMPALA/atari_agent.py
@@ -58,7 +58,10 @@ class AtariAgent(parl.Agent):
             lr = layers.data(
                 name='lr', shape=[1], dtype='float32', append_batch_size=False)
             entropy_coeff = layers.data(
-                name='entropy_coeff', shape=[], dtype='float32')
+                name='entropy_coeff',
+                shape=[1],
+                dtype='float32',
+                append_batch_size=False)
 
             self.learn_reader = fluid.layers.create_py_reader_by_data(
                 capacity=32,
diff --git a/examples/IMPALA/train.py b/examples/IMPALA/train.py
index bfb4ad5c0343a92a6179a44a9d5b6e1ecb9bc79b..cf9e55c54d1df8a14fc1751ac75303c6adab42ad 100755
--- a/examples/IMPALA/train.py
+++ b/examples/IMPALA/train.py
@@ -123,7 +123,7 @@ class Learner(object):
                     obs_np, actions_np, behaviour_logits_np, rewards_np,
                     dones_np,
                     np.float32(self.lr),
-                    np.float32(self.entropy_coeff)
+                    np.array([self.entropy_coeff], dtype='float32')
                 ]
 
     def run_learn(self):
diff --git a/examples/LiftSim_baseline/A2C/lift_agent.py b/examples/LiftSim_baseline/A2C/lift_agent.py
index 9c2d64d64c0f89dc939a2458d1b2e9fedbb5242a..1dd35e59a3c840b6468f5d95bcc6fccc07445bc8 100644
--- a/examples/LiftSim_baseline/A2C/lift_agent.py
+++ b/examples/LiftSim_baseline/A2C/lift_agent.py
@@ -67,7 +67,10 @@ class LiftAgent(parl.Agent):
             lr = layers.data(
                 name='lr', shape=[1], dtype='float32', append_batch_size=False)
             entropy_coeff = layers.data(
-                name='entropy_coeff', shape=[], dtype='float32')
+                name='entropy_coeff',
+                shape=[1],
+                dtype='float32',
+                append_batch_size=False)
 
             total_loss, pi_loss, vf_loss, entropy = self.alg.learn(
                 obs, actions, advantages, target_values, lr, entropy_coeff)
diff --git a/parl/algorithms/fluid/a3c.py b/parl/algorithms/fluid/a3c.py
index 27aa1dc8785315e4347cebcc3b13f8d80659e0a9..9b9f57e8eb5bfd59e3f79c1fc42e4d1374618f23 100644
--- a/parl/algorithms/fluid/a3c.py
+++ b/parl/algorithms/fluid/a3c.py
@@ -72,7 +72,6 @@ class A3C(Algorithm):
         policy_entropy = policy_distribution.entropy()
         entropy = layers.reduce_sum(policy_entropy)
 
-        entropy_coeff = layers.reshape(entropy_coeff, shape=[1])
         total_loss = (
             pi_loss + vf_loss * self.vf_loss_coeff + entropy * entropy_coeff)
 
diff --git a/parl/algorithms/fluid/impala/impala.py b/parl/algorithms/fluid/impala/impala.py
index bdfe53a67223165c13454260b59fc8e0a650b0e0..025f96f2650e3351552d6525c910d2f29406dbaa 100644
--- a/parl/algorithms/fluid/impala/impala.py
+++ b/parl/algorithms/fluid/impala/impala.py
@@ -78,7 +78,6 @@ class VTraceLoss(object):
         self.entropy = layers.reduce_sum(policy_entropy)
 
         # The summed weighted loss
-        entropy_coeff = layers.reshape(entropy_coeff, shape=[1])
         self.total_loss = (self.pi_loss + self.vf_loss * vf_loss_coeff +
                            self.entropy * entropy_coeff)