change: eliminate dependency on mnist dataset website (#986)

This commit will modify the folder structure to avoid needing to
download the datasets and relying on the dataset website from being
available.
This caused a canary to fail.

The dependency has existed since Pytorch 1.1 upgrade, as the directory
structure was modified in pytorch/vision#601

Author: knakad

tests/data/pytorch_mnist/mnist.py

@@ -47,7 +47,7 @@ def _get_train_data_loader(training_dir, is_distributed, batch_size, **kwargs):
         transform=transforms.Compose(
             [transforms.ToTensor(), transforms.Normalize((0.1307,), (0.3081,))]
         ),
-        download=True,
+        download=False,  # True sets a dependency on an external site for our canaries.
     )
     train_sampler = (
         torch.utils.data.distributed.DistributedSampler(dataset) if is_distributed else None
@@ -71,7 +71,7 @@ def _get_test_data_loader(training_dir, **kwargs):
             transform=transforms.Compose(
                 [transforms.ToTensor(), transforms.Normalize((0.1307,), (0.3081,))]
             ),
-            download=True,
+            download=False,  # True sets a dependency on an external site for our canaries.
         ),
         batch_size=1000,
         shuffle=True,

…pytorch_mnist/training/processed/test.pt …h_mnist/training/MNIST/processed/test.pttests/data/pytorch_mnist/training/processed/test.pt renamed to tests/data/pytorch_mnist/training/MNIST/processed/test.pt tests/data/pytorch_mnist/training/processed/test.pt renamed to tests/data/pytorch_mnist/training/MNIST/processed/test.pt

@@ -28,6 +28,8 @@
 from sagemaker.sklearn.model import SKLearnModel
 from tests.integ import DATA_DIR, PYTHON_VERSION
 
+MNIST_FOLDER_NAME = "MNIST"
+
 GIT_REPO = "https://github.com/aws/sagemaker-python-sdk.git"
 BRANCH = "test-branch-git-config"
 COMMIT = "ae15c9d7d5b97ea95ea451e4662ee43da3401d73"
@@ -69,7 +71,7 @@ def test_git_support_with_pytorch(sagemaker_local_session):
         git_config=git_config,
     )
 
-    pytorch.fit({"training": "file://" + os.path.join(data_path, "training")})
+    pytorch.fit({"training": "file://" + os.path.join(data_path, "training", MNIST_FOLDER_NAME)})
 
     with lock.lock(LOCK_PATH):
         try: