Update coach version for to 0.11.1 tensorflow. (aws#684)

Author: nadiaya

CHANGELOG.rst

@@ -15,6 +15,7 @@ CHANGELOG
 * bug-fix: pass accelerator_type in ``deploy`` for REST API TFS ``Model``
 * doc-fix: move content from tf/README.rst into sphynx project
 * doc-fix: Improve new developer experience in README
+* feature: Add support for Coach 0.11.1 for Tensorflow
 
 1.18.3.post1
 ============

src/sagemaker/rl/README.rst

@@ -4,7 +4,7 @@ SageMaker Reinforcement Learning Estimators
 
 With Reinforcement Learning (RL) Estimators, you can train reinforcement learning models on Amazon SageMaker.
 
-Supported versions of Coach: ``0.10.1`` with TensorFlow, ``0.11.0`` with TensorFlow or MXNet.
+Supported versions of Coach: ``0.11.1``, ``0.10.1`` with TensorFlow, ``0.11.0`` with TensorFlow or MXNet.
 For more information about Coach, see https://github.com/NervanaSystems/coach
 
 Supported versions of Ray: ``0.5.3`` with TensorFlow.
@@ -42,7 +42,7 @@ You can then create an ``RLEstimator`` with keyword arguments to point to this s
 
     rl_estimator = RLEstimator(entry_point='coach-train.py',
                                toolkit=RLToolkit.COACH,
-                               toolkit_version='0.11.0',
+                               toolkit_version='0.11.1',
                                framework=RLFramework.TENSORFLOW,
                                role='SageMakerRole',
                                train_instance_type='ml.p3.2xlarge',

src/sagemaker/rl/estimator.py

@@ -40,8 +40,11 @@
             'tensorflow': '1.11',
             'mxnet': '1.3'
         },
+        '0.11.1': {
+            'tensorflow': '1.12',
+        },
         '0.11': {
-            'tensorflow': '1.11',
+            'tensorflow': '1.12',
             'mxnet': '1.3'
         }
     },
@@ -69,7 +72,8 @@ class RLFramework(enum.Enum):
 class RLEstimator(Framework):
     """Handle end-to-end training and deployment of custom RLEstimator code."""
 
-    COACH_LATEST_VERSION = '0.11.0'
+    COACH_LATEST_VERSION_TF = '0.11.1'
+    COACH_LATEST_VERSION_MXNET = '0.11.0'
     RAY_LATEST_VERSION = '0.5.3'
 
     def __init__(self, entry_point, toolkit=None, toolkit_version=None, framework=None,

tests/conftest.py

@@ -39,8 +39,10 @@ def pytest_addoption(parser):
     parser.addoption('--mxnet-full-version', action='store', default=MXNet.LATEST_VERSION)
     parser.addoption('--ei-mxnet-full-version', action='store', default=MXNet.LATEST_VERSION)
     parser.addoption('--pytorch-full-version', action='store', default=PyTorch.LATEST_VERSION)
-    parser.addoption('--rl-coach-full-version', action='store',
-                     default=RLEstimator.COACH_LATEST_VERSION)
+    parser.addoption('--rl-coach-mxnet-full-version', action='store',
+                     default=RLEstimator.COACH_LATEST_VERSION_MXNET)
+    parser.addoption('--rl-coach-tf-full-version', action='store',
+                     default=RLEstimator.COACH_LATEST_VERSION_TF)
     parser.addoption('--rl-ray-full-version', action='store',
                      default=RLEstimator.RAY_LATEST_VERSION)
     parser.addoption('--sklearn-full-version', action='store', default=SKLEARN_VERSION)
@@ -128,7 +130,7 @@ def tf_version(request):
     return request.param
 
 
-@pytest.fixture(scope='module', params=['0.10.1', '0.10.1', '0.11', '0.11.0'])
+@pytest.fixture(scope='module', params=['0.10.1', '0.10.1', '0.11', '0.11.0', '0.11.1'])
 def rl_coach_tf_version(request):
     return request.param
 
@@ -164,8 +166,13 @@ def pytorch_full_version(request):
 
 
 @pytest.fixture(scope='module')
-def rl_coach_full_version(request):
-    return request.config.getoption('--rl-coach-full-version')
+def rl_coach_mxnet_full_version(request):
+    return request.config.getoption('--rl-coach-mxnet-full-version')
+
+
+@pytest.fixture(scope='module')
+def rl_coach_tf_full_version(request):
+    return request.config.getoption('--rl-coach-tf-full-version')
 
 
 @pytest.fixture(scope='module')

tests/integ/test_rl.py

@@ -27,8 +27,8 @@
 
 @pytest.mark.canary_quick
 @pytest.mark.skipif(PYTHON_VERSION != 'py3', reason="RL images supports only Python 3.")
-def test_coach_mxnet(sagemaker_session, rl_coach_full_version):
-    estimator = _test_coach(sagemaker_session, RLFramework.MXNET, rl_coach_full_version)
+def test_coach_mxnet(sagemaker_session, rl_coach_mxnet_full_version):
+    estimator = _test_coach(sagemaker_session, RLFramework.MXNET, rl_coach_mxnet_full_version)
 
     with timeout(minutes=15):
         estimator.fit(wait='False')
@@ -50,8 +50,8 @@ def test_coach_mxnet(sagemaker_session, rl_coach_full_version):
 
 
 @pytest.mark.skipif(PYTHON_VERSION != 'py3', reason="RL images supports only Python 3.")
-def test_coach_tf(sagemaker_session, rl_coach_full_version):
-    estimator = _test_coach(sagemaker_session, RLFramework.TENSORFLOW, rl_coach_full_version)
+def test_coach_tf(sagemaker_session, rl_coach_tf_full_version):
+    estimator = _test_coach(sagemaker_session, RLFramework.TENSORFLOW, rl_coach_tf_full_version)
 
     with timeout(minutes=15):
         estimator.fit()

tests/unit/test_rl.py

@@ -76,7 +76,7 @@ def _get_full_gpu_image_uri(toolkit, toolkit_version, framework):
 
 
 def _rl_estimator(sagemaker_session, toolkit=RLToolkit.COACH,
-                  toolkit_version=RLEstimator.COACH_LATEST_VERSION, framework=RLFramework.MXNET,
+                  toolkit_version=RLEstimator.COACH_LATEST_VERSION_MXNET, framework=RLFramework.MXNET,
                   train_instance_type=None, base_job_name=None, **kwargs):
     return RLEstimator(entry_point=SCRIPT_PATH,
                        toolkit=toolkit,
@@ -466,7 +466,7 @@ def test_wrong_framework_format(sagemaker_session):
 def test_wrong_toolkit_format(sagemaker_session):
     with pytest.raises(ValueError) as e:
         RLEstimator(toolkit='coach', framework=RLFramework.TENSORFLOW,
-                    toolkit_version=RLEstimator.COACH_LATEST_VERSION,
+                    toolkit_version=RLEstimator.COACH_LATEST_VERSION_TF,
                     entry_point=SCRIPT_PATH, role=ROLE, sagemaker_session=sagemaker_session,
                     train_instance_count=INSTANCE_COUNT, train_instance_type=INSTANCE_TYPE,
                     framework_version=None)