Merge branch 'master' into develop-action-buffer

.github/workflows/pytest.yml

         python -m pip install --progress-bar=off -r test_requirements.txt -c ${{ matrix.pip_constraints }}
         python -m pip install --progress-bar=off -e ./gym-unity -c ${{ matrix.pip_constraints }}
     - name: Save python dependencies
-      run: pip freeze > pip_versions-${{ matrix.python-version }}.txt
+      run: |
+        pip freeze > pip_versions-${{ matrix.python-version }}.txt
+        cat pip_versions-${{ matrix.python-version }}.txt
     - name: Run pytest
       run: pytest --cov=ml-agents --cov=ml-agents-envs --cov=gym-unity --cov-report html --junitxml=junit/test-results-${{ matrix.python-version }}.xml -p no:warnings
     - name: Upload pytest test results

Project/Assets/ML-Agents/Examples/SharedAssets/Scripts/ModelOverrider.cs

             var bp = m_Agent.GetComponent<BehaviorParameters>();
             var behaviorName = bp.BehaviorName;
 
-            var nnModel = GetModelForBehaviorName(behaviorName);
+            NNModel nnModel = null;
+            try
+            {
+                nnModel = GetModelForBehaviorName(behaviorName);
+            }
+            catch (Exception e)
+            {
+                overrideError = $"Exception calling GetModelForBehaviorName: {e}";
+            }
+
-                overrideError =
-                    $"Didn't find a model for behaviorName {behaviorName}. Make " +
-                    $"sure the behaviorName is set correctly in the commandline " +
-                    $"and that the model file exists";
+                if (string.IsNullOrEmpty(overrideError))
+                {
+                    overrideError =
+                        $"Didn't find a model for behaviorName {behaviorName}. Make " +
+                        "sure the behaviorName is set correctly in the commandline " +
+                        "and that the model file exists";
+                }
             }
             else
             {

com.unity.ml-agents/CHANGELOG.md

 #### com.unity.ml-agents (C#)
 #### ml-agents / ml-agents-envs / gym-unity (Python)
  - PyTorch trainers are now the default. See the
- [installation docs](https://github.com/Unity-Technologies/ml-agents/blob/mastere/docs/Installation.md) for
+ [installation docs](https://github.com/Unity-Technologies/ml-agents/blob/master/docs/Installation.md) for
  more information on installing PyTorch. For the time being, TensorFlow is still available;
  you can use the TensorFlow backend by adding `--tensorflow` to the CLI, or
  adding `framework: tensorflow` in the configuration YAML. (#4517)
 - The Barracuda dependency was upgraded to 1.1.2 (#4571)
 #### ml-agents / ml-agents-envs / gym-unity (Python)
+- The `action_probs` node is no longer listed as an output in TensorFlow models (#4613).
 
 ### Bug Fixes
 #### com.unity.ml-agents (C#)

docs/Installation.md

 installing ML-Agents. Activate your virtual environment and run from the command line:
 
 ```sh
-pip3 install torch -f https://download.pytorch.org/whl/torch_stable.html
+pip3 install torch==1.7.0 -f https://download.pytorch.org/whl/torch_stable.html
 ```
 
 Note that on Windows, you may also need Microsoft's

ml-agents/mlagents/trainers/tests/torch/test_reward_providers/test_gail.py

         RewardSignalType.GAIL, behavior_spec, gail_settings
     )
 
-    for _ in range(200):
+    for _ in range(300):
         gail_rp.update(buffer_policy)
         reward_expert = gail_rp.evaluate(buffer_expert)[0]
         reward_policy = gail_rp.evaluate(buffer_policy)[0]

ml-agents/mlagents/trainers/tests/torch/test_reward_providers/utils.py

 ) -> AgentBuffer:
     buffer = AgentBuffer()
     curr_observations = [
-        np.random.normal(size=shape) for shape in behavior_spec.observation_shapes
+        np.random.normal(size=shape).astype(np.float32)
+        for shape in behavior_spec.observation_shapes
-        np.random.normal(size=shape) for shape in behavior_spec.observation_shapes
+        np.random.normal(size=shape).astype(np.float32)
+        for shape in behavior_spec.observation_shapes
     ]
     action_buffer = behavior_spec.action_spec.random_action(1)
     action = {}

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

 
 def test_list_to_tensor():
     # Test converting pure list
-    unconverted_list = [[1, 2], [1, 3], [1, 4]]
+    unconverted_list = [[1.0, 2], [1, 3], [1, 4]]
     tensor = ModelUtils.list_to_tensor(unconverted_list)
     # Should be equivalent to torch.tensor conversion
     assert torch.equal(tensor, torch.tensor(unconverted_list))
     list_of_np = [np.asarray(_el) for _el in unconverted_list]
     tensor = ModelUtils.list_to_tensor(list_of_np)
     # Should be equivalent to torch.tensor conversion
-    assert torch.equal(tensor, torch.tensor(unconverted_list))
+    assert torch.equal(tensor, torch.tensor(unconverted_list, dtype=torch.float32))
 
 
 def test_break_into_branches():

ml-agents/mlagents/trainers/tf/model_serialization.py

     ]
 )
 
-POSSIBLE_OUTPUT_NODES = frozenset(
-    ["action", "action_probs", "recurrent_out", "value_estimate"]
-)
+POSSIBLE_OUTPUT_NODES = frozenset(["action", "recurrent_out", "value_estimate"])
 
 MODEL_CONSTANTS = frozenset(
     [

ml-agents/mlagents/trainers/torch/distributions.py

         if self.conditional_sigma:
             log_sigma = torch.clamp(self.log_sigma(inputs), min=-20, max=2)
         else:
-            # Expand so that entropy matches batch size
-            log_sigma = self.log_sigma.expand(inputs.shape[0], -1)
+            # Expand so that entropy matches batch size. Note that we're using
+            # torch.cat here instead of torch.expand() becuase it is not supported in the
+            # verified version of Barracuda (1.0.2).
+            log_sigma = torch.cat([self.log_sigma] * inputs.shape[0], axis=0)
         if self.tanh_squash:
             return [TanhGaussianDistInstance(mu, torch.exp(log_sigma))]
         else:

ml-agents/mlagents/trainers/torch/utils.py

 
     @staticmethod
     def list_to_tensor(
-        ndarray_list: List[np.ndarray], dtype: Optional[torch.dtype] = None
+        ndarray_list: List[np.ndarray], dtype: Optional[torch.dtype] = torch.float32
     ) -> torch.Tensor:
         """
         Converts a list of numpy arrays into a tensor. MUCH faster than

ml-agents/setup.py

         "Pillow>=4.2.1",
         "protobuf>=3.6",
         "pyyaml>=3.1.0",
-        # Windows ver. of PyTorch doesn't work from PyPi
-        'torch>=1.6.0;platform_system!="Windows"',
+        # Windows ver. of PyTorch doesn't work from PyPi. Installation:
+        # https://github.com/Unity-Technologies/ml-agents/blob/master/docs/Installation.md#windows-installing-pytorch
+        'torch>=1.6.0,<1.8.0;platform_system!="Windows"',
-        "cattrs>=1.0.0",
+        # cattrs 1.1.0 dropped support for python 3.6.
+        "cattrs>=1.0.0,<1.1.0",
         "attrs>=19.3.0",
         'pypiwin32==223;platform_system=="Windows"',
     ],