Filter empty wikipedia pages after cleaning.

PiperOrigin-RevId: 280464382

Author: adarob

tensorflow_datasets/text/wikipedia.py

@@ -116,7 +116,7 @@ class Wikipedia(tfds.core.BeamBasedBuilder):
   BUILDER_CONFIGS = [
       WikipediaConfig(  # pylint:disable=g-complex-comprehension
           version=tfds.core.Version(
-              "0.0.3", experiments={tfds.core.Experiment.S3: False}),
+              "0.0.4", experiments={tfds.core.Experiment.S3: False}),
           language=lang,
           date="20190301",
       ) for lang in WIKIPEDIA_LANGUAGES
@@ -224,6 +224,10 @@ def _clean_content(inputs):
         logging.error("mwparserfromhell ParseError: %s", e)
         return
 
+      if not text:
+        beam.metrics.Metrics.counter(language, "empty-clean-examples").inc()
+        return
+
       beam.metrics.Metrics.counter(language, "cleaned-examples").inc()
 
       yield {