Fix feature handling in Radon and NaN checks in LAION400M.

In radon_dataset_builder.py, the feature extraction now iterates through the defined features and includes error handling for conversion failures.
In laion400m.py, NaN values for similarity and license are now checked using pd.isna() for more robust handling.

PiperOrigin-RevId: 911744580

Author: tomvdw

tensorflow_datasets/datasets/radon/radon_dataset_builder.py

@@ -133,9 +133,15 @@ def _generate_examples(self, file_path_srrs2, file_path_cty):
 
     for i, (_, row) in enumerate(df.iterrows()):
       radon_val = row.pop('activity')
+      features_dict = {}
+      for name, (_, convert_fn) in features().items():
+        try:
+          features_dict[name] = convert_fn(row[name])
+        except ValueError as e:
+          raise ValueError(
+              f'Failed to convert {name} with value {row[name]!r}'
+          ) from e
       yield i, {
           'activity': float(radon_val),
-          'features': {
-              name: features()[name][1](value) for name, value in row.items()
-          },
+          'features': features_dict,
       }

tensorflow_datasets/vision_language/laion400m/laion400m.py

@@ -14,12 +14,12 @@
 # limitations under the License.
 
 """LAION-400M image dataset."""
+
 import functools
 from typing import Dict, Tuple
 
 from etils import epath
 import numpy as np
-
 from tensorflow_datasets.core.utils.lazy_imports_utils import tensorflow as tf
 import tensorflow_datasets.public_api as tfds
 
@@ -78,15 +78,21 @@
 
 def _get_example_metadata(metadata_df_row):
   """Returns example metadata."""
+  pd = tfds.core.lazy_imports.pandas
   nsfw_tag = metadata_df_row['NSFW']
   if nsfw_tag not in _NSFW_TAGS:
     nsfw_tag = _NSFW_MISSING_TAG
 
+  similarity = metadata_df_row['similarity']
+  license_ = metadata_df_row['LICENSE']
+
   return {
       'caption': metadata_df_row['caption'],
       'nsfw': nsfw_tag,
-      'similarity': metadata_df_row['similarity'] or _MISSING_SIMILARITY_VALUE,
-      'license': metadata_df_row['LICENSE'] or '',
+      'similarity': (
+          _MISSING_SIMILARITY_VALUE if pd.isna(similarity) else similarity
+      ),
+      'license': '' if pd.isna(license_) else license_,
       'url': metadata_df_row['url'],
       'original_width': metadata_df_row['original_width'],
       'original_height': metadata_df_row['original_height'],
@@ -174,11 +180,9 @@ def _info(self) -> tfds.core.DatasetInfo:
     }
 
     if self.builder_config.name == LAION400M_IMAGES_CONFIG.name:
-      features.update(
-          {
-              'image': tfds.features.Image(doc='image'),
-          }
-      )
+      features.update({
+          'image': tfds.features.Image(doc='image'),
+      })
     else:
       features.update({
           'image_embedding': tfds.features.Tensor(