fixed torch test sac

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

         cont_sampled_actions = sampled_actions.continuous_tensor
 
         cont_actions = actions.continuous_tensor
-        disc_actions = actions.discrete_tensor
         q1p_out, q2p_out = self.value_network(
             vec_obs,
             vis_obs,
             sequence_length=self.policy.sequence_length,
         )
 
-        if self._action_spec.discrete_size:
+        if self._action_spec.discrete_size > 0:
+            disc_actions = actions.discrete_tensor
             q1_stream = self._condense_q_streams(q1_out, disc_actions)
             q2_stream = self._condense_q_streams(q2_out, disc_actions)
         else:

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

         "Losses/Value Loss",
         "Losses/Q1 Loss",
         "Losses/Q2 Loss",
-        "Policy/Entropy Coeff",
+        "Policy/Continuous Entropy Coeff",
+        "Policy/Discrete Entropy Coeff",
         "Policy/Learning Rate",
     ]
     for stat in required_stats:

ml-agents/mlagents/trainers/tests/torch/test_simple_rl.py

 
 @pytest.mark.parametrize("action_sizes", [(0, 1), (1, 0)])
 def test_recurrent_sac(action_sizes):
-    step_size = 0.2 if action_sizes else 0.5
+    step_size = 0.2 if action_sizes == (0, 1) else 0.5
     env = MemoryEnvironment(
         [BRAIN_NAME], action_sizes=action_sizes, step_size=step_size
     )