Fixed inconsistent action placeholder (#5213)

python/ray/rllib/agents/marwil/marwil_policy.py

@@ -110,7 +110,7 @@ class MARWILPolicy(MARWILPostprocessing, TFPolicy):
         self.output_actions = action_dist.sample()
 
         # Training inputs
-        self.act_t = tf.placeholder(tf.int32, [None], name="action")
+        self.act_t = ModelCatalog.get_action_placeholder(action_space)
         self.cum_rew_t = tf.placeholder(tf.float32, [None], name="reward")
 
         # v network evaluation