Slight speedup

nbaiot.py

@@ -105,33 +105,31 @@ class NBAIOTDataset(datasets.GeneratorBasedBuilder):
     def _generate_examples(self, filepath, split):
         for device in _DEVICE_NAMES:
             # First load in the benign traffic
-            all_data = pd.read_csv(f"{filepath}/{device}/benign_traffic.csv")
-            all_data['attack'] = "benign_traffic"
+            all_data = pd.read_csv(f"{filepath}/{device}/benign_traffic.csv").values
+            attacks = np.repeat("benign_traffic", len(all_data))
             # Then the standard attacks
             attacks_rar = rarfile.RarFile(f"{filepath}/{device}/gafgyt_attacks.rar")
             for fileinfo in attacks_rar.infolist():
                 with attacks_rar.open(fileinfo.filename) as f:
-                    df = pd.read_csv(f)
-                df['attack'] = f.name.replace(".csv", "")
-                all_data = pd.concat((all_data, df))
+                    data = pd.read_csv(f).values
+                attacks = np.concatenate((attacks, np.repeat(f.name.replace(".csv", ""), len(data))))
+                all_data = np.concatenate((all_data, data))
             # And, if present, the Mirai attacks
             if device not in ["Ennio_Doorbell", "Samsung_SNH_1011_N_Webcam"]:
                 mirai_rar = rarfile.RarFile(f"{filepath}/{device}/mirai_attacks.rar")
                 for fileinfo in mirai_rar.infolist():
                     with mirai_rar.open(fileinfo.filename) as f:
-                        df = pd.read_csv(f)
-                    df['attack'] = "mirai-" + fileinfo.filename.replace(".csv", "")
-                    all_data = pd.concat((all_data, df))
+                        data = pd.read_csv(f).values
+                    attacks = np.concatenate((attacks, np.repeat("mirai-" + f.name.replace(".csv", ""), len(data))))
+                    all_data = np.concatenate((all_data, data))
             # Create the train-test split
             rng = np.random.default_rng(round(np.pi**(np.pi * 100)))
             train = rng.uniform(size=len(all_data)) < 0.85
-            all_data = all_data.iloc[np.where(train if split == "train" else ~train)]
-            attacks = all_data['attack'].to_list()
-            all_data = all_data.drop(columns="attack")
+            all_data = all_data[train if split == "train" else ~train]
             # Finally yield the data
-            for (key, row), attack in zip(all_data.iterrows(), attacks):
+            for key, (data, attack) in enumerate(zip(all_data, attacks)):
                 yield key, {
-                    "features": row.to_numpy(),
+                    "features": data,
                     "attack": attack,
                     "device": device,
                 }