Add optional burn-in for SAC as well

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

     def construct_feed_dict(
         self, mini_batch: AgentBuffer, num_sequences: int
     ) -> Dict[tf.Tensor, Any]:
-        # Do a burn-in for memories
+        # Do an optional burn-in for memories
         num_burn_in = int(BURN_IN_RATIO * self.policy.sequence_length)
         burn_in_mask = np.ones((self.policy.sequence_length), dtype=np.float32)
         burn_in_mask[range(0, num_burn_in)] = 0

ml-agents/mlagents/trainers/sac/optimizer.py

 POLICY_SCOPE = ""
 TARGET_SCOPE = "target_network"
 
+BURN_IN_RATIO = 0.0
+
 
 class SACOptimizer(TFOptimizer):
     def __init__(self, policy: TFPolicy, trainer_params: Dict[str, Any]):
         :param batch: Mini-batch to use to update.
         :param num_sequences: Number of LSTM sequences in batch.
         """
+        # Do an optional burn-in for memories
+        num_burn_in = int(BURN_IN_RATIO * self.policy.sequence_length)
+        burn_in_mask = np.ones((self.policy.sequence_length), dtype=np.float32)
+        burn_in_mask[range(0, num_burn_in)] = 0
+        burn_in_mask = np.tile(burn_in_mask, num_sequences)
-            self.policy.mask_input: batch["masks"],
+            self.policy.mask_input: batch["masks"] * burn_in_mask,
         }
         for name in self.reward_signals:
             feed_dict[self.rewards_holders[name]] = batch["{}_rewards".format(name)]