Refactor script

top_quark_tagging.py

@@ -44,11 +44,11 @@ _URLS = {
         "test-labels": "data/test-labels.parquet",
     },
     "image": {
-        "train": "data/train-images.parquet",
+        "train": "data/train-images.zip",
         "train-labels": "data/train-labels.parquet",
-        "validation": "data/train-images.parquet",
+        "validation": "data/train-images.zip",
         "validation-labels": "data/validation-labels.parquet",
-        "test": "data/train-images.parquet",
+        "test": "data/train-images.zip",
         "test-labels": "data/test-labels.parquet",
     },
 }
@@ -890,7 +890,7 @@ class TopQuarkTagging(datasets.GeneratorBasedBuilder):
             features = datasets.Features(features_data)
         else:
             features = datasets.Features(
-                {"image": datasets.Image(), "is_signal_new": datasets.ClassLabel(names=["qcd", "top"])}
+                {"file": datasets.Value("string"), "is_signal_new": datasets.ClassLabel(names=["qcd", "top"])}
             )
         return datasets.DatasetInfo(
             description=_DESCRIPTION,
@@ -923,12 +923,17 @@ class TopQuarkTagging(datasets.GeneratorBasedBuilder):
 
     def _generate_examples(self, filepath, split):
         """Generate examples."""
-        examples = datasets.load_dataset("parquet", data_files={split: filepath[0]}, split=split)
-        labels = datasets.load_dataset("parquet", data_files={split: filepath[1]}, split=split)
-        for id_, (row, label) in enumerate(zip(examples, labels)):
-            if self.config.name == "image":
-                values = np.fromiter(row.values(), dtype=float)
-                image = values.reshape(33, 33)
-                yield id_, {"image": image, **label}
-            else:
+        if self.config.name == "image":
+            examples = pd.read_csv(f"{filepath[0]}/train-images/train.csv")
+            for row in examples.itertuples():
+                yield row.Index, {"file": f"{filepath[0]}/{row.file}", "is_signal_new": row.label}
+        else:
+            examples = datasets.load_dataset("parquet", data_files={split: filepath[0]}, split=split)
+            labels = datasets.load_dataset("parquet", data_files={split: filepath[1]}, split=split)
+            for id_, (row, label) in enumerate(zip(examples, labels)):
+                # if self.config.name == "image":
+                #     values = np.fromiter(row.values(), dtype=float)
+                #     image = values.reshape(33, 33)
+                #     yield id_, {"image": image, **label}
+                # else:
                 yield id_, {**row, **label}