upd: I don't know anymore

rwth_phoenix_weather_2014.py

@@ -79,7 +79,7 @@ class RWTHPhoenixWeather2014(datasets.GeneratorBasedBuilder):
 
     def _info(self):
         features_dict = {
-            "tokens": datasets.Sequence(feature=datasets.Value(dtype='string')),
+            "transcription": datasets.Value("string"),
         }
 
         if self.config.name != "pre-training":
@@ -97,8 +97,8 @@ class RWTHPhoenixWeather2014(datasets.GeneratorBasedBuilder):
 
     def _split_generators(self, dl_manager: datasets.DownloadManager):
         split_frames = {}
-        split_tokens = {}
-        split_ids = {}
+        split_transcription = {}
+        file_ids = {}
 
         dataDirMapper = {
             datasets.Split.TRAIN: "train",
@@ -117,19 +117,16 @@ class RWTHPhoenixWeather2014(datasets.GeneratorBasedBuilder):
                 f"{base_url}/annotations/manual/{dataDirMapper[split]}{self.config.corpus_file_suffix}")
             df = pd.read_csv(data_csv, sep='|')
 
-            split_tokens[split] = df.annotation.map(
-                lambda x: x.strip().split()
-            )
-
-            split_ids[split] = df.id
+            split_transcription[split] = df['annotation']
+            file_ids[split] = df['id']
 
             if self.config.name == 'pre-training':
                 split_frames[split] = [None for id in df.id]
                 continue
 
             frame_archive_urls = dl_manager.download([
-                f"{base_url}/features/fullFrame-210x260px/{dataDirMapper[split]}/{id}.tar"
-                for id in df.id
+                f"{base_url}/features/fullFrame-210x260px/{dataDirMapper[split]}/{file_id}.tar"
+                for file_id in file_ids[split]
             ])
 
             split_frames[split] = [
@@ -141,9 +138,10 @@ class RWTHPhoenixWeather2014(datasets.GeneratorBasedBuilder):
             datasets.SplitGenerator(
                 name=split,
                 gen_kwargs={
-                    "split_ids": split_ids[split],
+                    "split_name": split,
+                    "file_ids": file_ids[split],
                     "split_frames": split_frames[split],
-                    "split_tokens": split_tokens[split],
+                    "split_transcription": split_transcription[split],
                 },
             )
             for split in [
@@ -153,17 +151,17 @@ class RWTHPhoenixWeather2014(datasets.GeneratorBasedBuilder):
             ]
         ]
 
-    def _generate_examples(self, split_ids, split_frames, split_tokens):
-        for id, frames, tokens in zip(split_ids, split_frames, split_tokens):
+    def _generate_examples(self, split_name, file_ids, split_frames, split_transcription):
+        for file_id, frames, transcription in zip(file_ids, split_frames, split_transcription):
             frames_as_list = [
                 {"path": p, "bytes": im.read()} for p, im in frames
             ] if frames is not None else None
 
             result = {
-                "tokens": tokens,
+                "transcription": transcription,
             }
 
             if frames_as_list is not None:
                 result["frames"] = frames_as_list
 
-            yield id, result
+            yield f"{split_name}/{file_id}", result