resolve the compatibility issue (#226)

* fix compatibility issue with the newest paddle

* remove logging lines

* resolve the compatibility issue with the newest paddle

* yapf
Co-authored-by: Nrobot <zenghongsheng@baidu.com>

examples/DQN/atari_agent.py

@@ -106,7 +106,7 @@ class AtariAgent(parl.Agent):
             'reward': reward,
             'next_obs': next_obs.astype('float32'),
             'terminal': terminal,
-            'lr': lr
+            'lr': np.float32(lr)
         }
         cost = self.fluid_executor.run(
             self.learn_program, feed=feed, fetch_list=[self.cost])[0]

examples/IMPALA/train.py

@@ -121,7 +121,9 @@ class Learner(object):
 
                 yield [
                     obs_np, actions_np, behaviour_logits_np, rewards_np,
-                    dones_np, self.lr, self.entropy_coeff
+                    dones_np,
+                    np.float32(self.lr),
+                    np.float32(self.entropy_coeff)
                 ]
 
     def run_learn(self):

parl/algorithms/fluid/impala/impala.py

@@ -78,6 +78,7 @@ 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)