Remove random normal epsilon (#3039)

ml-agents/mlagents/trainers/ppo/policy.py

             feed_dict[self.model.epsilon] = epsilon
         feed_dict = self.fill_eval_dict(feed_dict, brain_info)
         run_out = self._execute_model(feed_dict, self.inference_dict)
-        if self.use_continuous_act:
-            run_out["random_normal_epsilon"] = epsilon
         return run_out
 
     @timed
 
         if self.use_continuous_act:
             feed_dict[model.output_pre] = mini_batch["actions_pre"]
-            feed_dict[model.epsilon] = mini_batch["random_normal_epsilon"]
         else:
             feed_dict[model.action_holder] = mini_batch["actions"]
             if self.use_recurrent:

ml-agents/mlagents/trainers/ppo/trainer.py

             self.processing_buffer[agent_id]["actions_pre"].append(
                 actions_pre[agent_idx]
             )
-            epsilons = take_action_outputs["random_normal_epsilon"]
-            self.processing_buffer[agent_id]["random_normal_epsilon"].append(
-                epsilons[agent_idx]
-            )
         a_dist = take_action_outputs["log_probs"]
         # value is a dictionary from name of reward to value estimate of the value head
         self.processing_buffer[agent_id]["actions"].append(actions[agent_idx])

ml-agents/mlagents/trainers/tests/mock_brain.py

         buffer[0]["actions_pre"].append(
             np.ones(buffer[0]["actions"][0].shape, dtype=np.float32)
         )
-        buffer[0]["random_normal_epsilon"].append(
-            np.ones(buffer[0]["actions"][0].shape, dtype=np.float32)
-        )
         buffer[0]["action_mask"].append(
             np.ones(np.sum(brain_params.vector_action_space_size), dtype=np.float32)
         )

ml-agents/mlagents/trainers/tests/test_sac.py

 
     # Test update, while removing PPO-specific buffer elements.
     update_buffer = mb.simulate_rollout(
-        env,
-        policy,
-        BUFFER_INIT_SAMPLES,
-        exclude_key_list=["advantages", "actions_pre", "random_normal_epsilon"],
+        env, policy, BUFFER_INIT_SAMPLES, exclude_key_list=["advantages", "actions_pre"]
     )
 
     # Mock out reward signal eval