Be more explicit on how to use tar datasets with DataLoader in our ex…

…ample (#103)

Co-authored-by: Simon Beal <[email protected]>

examples/Getting started with the Amazon S3 Connector for PyTorch.ipynb

@@ -643,7 +643,7 @@
     {
      "data": {
       "text/plain": [
-       "'2022.001.0000.00.DISC.01.jpg.image.jpg'"
+       "['2022.001.0000.00.DISC.01.jpg.image.jpg']"
       ]
      },
      "metadata": {},
@@ -663,7 +663,7 @@
     {
      "data": {
       "text/plain": [
-       "'2022.001.0320.00.DISC.01.jpg.image.jpg'"
+       "['2022.001.0320.00.DISC.01.jpg.image.jpg']"
       ]
      },
      "metadata": {},
@@ -683,7 +683,7 @@
     {
      "data": {
       "text/plain": [
-       "'2022.001.0640.00.DISC.01.jpg.image.jpg'"
+       "['2022.001.0640.00.DISC.01.jpg.image.jpg']"
       ]
      },
      "metadata": {},
@@ -745,12 +745,15 @@
     "def shard_to_dict(object):\n",
     "    return {\"url\": object.key, \"stream\": object}\n",
     "\n",
-    "dataset = s3torchconnector.S3IterableDataset.from_prefix(SHARDS_URI, region=REGION, transform=shard_to_dict)\n",
-    "dataset = webdataset.tariterators.tar_file_expander(dataset)\n",
+    "s3_dataset = s3torchconnector.S3IterableDataset.from_prefix(SHARDS_URI, region=REGION, transform=shard_to_dict)\n",
+    "tar_dataset = webdataset.tariterators.tar_file_expander(s3_dataset)\n",
+    "dataset = torchdata.datapipes.iter.IterableWrapper(tar_dataset, deepcopy=False)\n",
     "\n",
-    "for sample in itertools.islice(dataset, 0, 100, 20):\n",
+    "loader = torch.utils.data.DataLoader(dataset)\n",
+    "\n",
+    "for sample in itertools.islice(loader, 0, 100, 20):\n",
     "    display(sample[\"fname\"])\n",
-    "    display(Image.open(io.BytesIO(sample[\"data\"])).reduce(8))"
+    "    display(Image.open(io.BytesIO(sample[\"data\"][0])).reduce(8))"
    ]
   },
   {