Use the test argement framework_version in all tests (aws#158)

* Use the test argement framework_version in all tests

* Make flake8 happy

Author: icywang86rui

test/integration/local/test_keras.py

@@ -16,99 +16,42 @@
 import os
 
 import numpy as np
-<<<<<<< HEAD
-<<<<<<< HEAD
-=======
->>>>>>> Fix Keras test (#132)
 import pytest
 from sagemaker.tensorflow import serving, TensorFlow
 
 from test.integration import RESOURCE_PATH
 from test.integration.utils import processor, py_version  # noqa: F401
-=======
-from sagemaker.tensorflow import serving, TensorFlow
-
-from test.integration import RESOURCE_PATH
->>>>>>> Add Keras support (#126)
 
 
 logging.basicConfig(level=logging.DEBUG)
 
 
-<<<<<<< HEAD
-<<<<<<< HEAD
-<<<<<<< HEAD
 @pytest.mark.skip(reason="Serving part fails because of version mismatch.")
 def test_keras_training(sagemaker_local_session, docker_image, tmpdir, framework_version):
-=======
-def test_keras_training(sagemaker_local_session, docker_image, tmpdir):
->>>>>>> Add Keras support (#126)
-=======
-@pytest.fixture
-def local_mode_instance_type(processor):
-    instance_type = 'local' if processor == 'cpu' else 'local_gpu'
-    return instance_type
-
-
-def test_keras_training(sagemaker_local_session, docker_image, tmpdir, local_mode_instance_type):
->>>>>>> Fix Keras test (#132)
-=======
-@pytest.mark.skip_gpu
-def test_keras_training(sagemaker_local_session, docker_image, tmpdir):
->>>>>>> Skip keras local mode test on gpu and use random port for serving in the test (#134)
     entry_point = os.path.join(RESOURCE_PATH, 'keras_inception.py')
     output_path = 'file://{}'.format(tmpdir)
 
     estimator = TensorFlow(
         entry_point=entry_point,
         role='SageMakerRole',
         train_instance_count=1,
-<<<<<<< HEAD
-<<<<<<< HEAD
-        train_instance_type='local',
-<<<<<<< HEAD
-<<<<<<< HEAD
-=======
-        train_instance_type=local_mode_instance_type,
->>>>>>> Fix Keras test (#132)
-=======
         train_instance_type='local',
->>>>>>> Skip keras local mode test on gpu and use random port for serving in the test (#134)
         image_name=docker_image,
         sagemaker_session=sagemaker_local_session,
         model_dir='/opt/ml/model',
         output_path=output_path,
         framework_version=framework_version,
-=======
-=======
-        image_name=docker_image,
->>>>>>> Create parameter server in different thread (#129)
-        sagemaker_session=sagemaker_local_session,
-        model_dir='/opt/ml/model',
-        output_path=output_path,
-        framework_version='1.11.0',
->>>>>>> Add Keras support (#126)
         py_version='py3')
 
     estimator.fit()
 
-<<<<<<< HEAD
-<<<<<<< HEAD
     model = serving.Model(model_data=output_path,
                           role='SageMakerRole',
                           framework_version=framework_version,
-=======
-    model = serving.Model(model_data=output_path, role='SageMakerRole',
-=======
-    model = serving.Model(model_data=output_path,
-                          role='SageMakerRole',
->>>>>>> Create parameter server in different thread (#129)
-                          framework_version='1.11.0',
->>>>>>> Add Keras support (#126)
                           sagemaker_session=sagemaker_local_session)
 
     predictor = model.deploy(initial_instance_count=1, instance_type='local')
 
     assert predictor.predict(np.random.randn(4, 4, 4, 2) * 255)
 
-    predictor.delete_endpoint()
+    predictor.delete_endpoint()