For PyTorch samples (with and without Horovod), the manner in which...

For PyTorch samples (with and without Horovod), the manner in which pre-downloaded datasets are loaded has been changed a bit to comply with the versions of torch and torchvision installed in stage 2019a.

horovod/pytorch/mnist.py

@@ -57,7 +57,7 @@ if args.cuda:
 dataset_file = os.path.join(data_dir, data_file)
 
 # [HPCNS] Dataset filename for this rank
-dataset_for_rank = 'MNIST-data-%d' % hvd.rank()
+dataset_for_rank = 'MNIST'
 
 # [HPCNS] If the path already exists, remove it
 if os.path.exists(dataset_for_rank):
@@ -68,7 +68,7 @@ shutil.copytree(dataset_file, dataset_for_rank)
 
 kwargs = {'num_workers': 1, 'pin_memory': True} if args.cuda else {}
 train_dataset = \
-    datasets.MNIST(dataset_for_rank, train=True, download=False,
+    datasets.MNIST('', train=True, download=False,
                    transform=transforms.Compose([
                        transforms.ToTensor(),
                        transforms.Normalize((0.1307,), (0.3081,))
@@ -80,7 +80,7 @@ train_loader = torch.utils.data.DataLoader(
     train_dataset, batch_size=args.batch_size, sampler=train_sampler, **kwargs)
 
 test_dataset = \
-    datasets.MNIST(dataset_for_rank, train=False, download=False, transform=transforms.Compose([
+    datasets.MNIST('', train=False, download=False, transform=transforms.Compose([
         transforms.ToTensor(),
         transforms.Normalize((0.1307,), (0.3081,))
     ]))

pytorch/mnist.py

@@ -108,7 +108,7 @@ def main():
     dataset_file = os.path.join(data_dir, data_file)
 
     # [HPCNS] A copy of the dataset in the current directory
-    dataset_copy = 'MNIST-data'
+    dataset_copy = 'MNIST'
 
     # [HPCNS] If the path already exists, remove it
     if os.path.exists(dataset_copy):
@@ -120,14 +120,14 @@ def main():
 
     kwargs = {'num_workers': 1, 'pin_memory': True} if use_cuda else {}
     train_loader = torch.utils.data.DataLoader(
-        datasets.MNIST(dataset_copy, train=True, download=False,
+        datasets.MNIST('', train=True, download=False,
                        transform=transforms.Compose([
                            transforms.ToTensor(),
                            transforms.Normalize((0.1307,), (0.3081,))
                        ])),
         batch_size=args.batch_size, shuffle=True, **kwargs)
     test_loader = torch.utils.data.DataLoader(
-        datasets.MNIST(dataset_copy, train=False, download=False, transform=transforms.Compose([
+        datasets.MNIST('', train=False, download=False, transform=transforms.Compose([
             transforms.ToTensor(),
             transforms.Normalize((0.1307,), (0.3081,))
         ])),