tensorflow/models · error · ValueError

Unknown feature source

Error message

Unknown feature source {self._feature_sources[i]} for {name}

What it means

Per-feature validation in the YT8M input parser's __init__: while building tf.io parse specs, feature_sources[i] must be one of the recognized sources — 'feature' (sequence-level, FixedLenSequenceFeature), 'context' (FixedLenFeature), or label sources like 'ids'/'rgb'/'audio'. The value at index i matched none, so the parser cannot decide how to declare that feature in the example spec; the message names the offending source value and feature name.

Solutions

1. Set each entry of feature_sources to a supported value: 'feature', 'context', 'ids', 'rgb', or 'audio'
2. Align feature_sources, feature_names, feature_sizes, and feature_dtypes element-wise — check for truncation or reordering
3. Check for typos like 'features' or 'Context'

When it happens

Trigger: Thrown at official/projects/yt8m/dataloaders/yt8m_input.py:303 when the library encounters an invalid state.

Common situations: See trigger scenarios.

AI-assisted analysis of tensorflow/models@e006f5f0d5 (2026-08-24). Data as JSON: /api/errors/7a00b23abb50c5be. Report an issue: GitHub.

Appendix: source

Thrown at official/projects/yt8m/dataloaders/yt8m_input.py:303

    else:
      self._add_labels_specification()

    for i, name in enumerate(self._feature_names):
      if self._feature_from_bytes[i]:
        feature_type = tf.io.FixedLenSequenceFeature([], dtype=tf.string)
      else:
        dtype = tf.dtypes.as_dtype(self._feature_dtypes[i])
        feature_shape = [self._feature_sizes[i]]
        if self._feature_sources[i] == "feature":
          feature_type = tf.io.FixedLenSequenceFeature(feature_shape, dtype)
        else:
          feature_type = tf.io.FixedLenFeature(feature_shape, dtype)
      if self._feature_sources[i] == "feature":
        self._sequence_features[name] = feature_type
      elif self._feature_sources[i] == "context":
        self._context_features[name] = feature_type
      else:
        raise ValueError(
            f"Unknown feature source {self._feature_sources[i]} for {name}")

  def _add_labels_specification(self):
    if not self._label_field:
      raise ValueError(f"Invalid label field: {self._label_field}!")
    self._context_features.update(
        {self._label_field: tf.io.VarLenFeature(tf.int64)})

  def decode(self,
             serialized_example: tf.train.SequenceExample) -> Dict[str, Any]:
    """Parses a single tf.train.SequenceExample into video and label tensors."""
    contexts, features = tf.io.parse_single_sequence_example(
        serialized_example,
        context_features=self._context_features,
        sequence_features=self._sequence_features)
    decoded_tensor = {**contexts, **features}
    for i, name in enumerate(self._feature_names):
      # Convert the VarLen feature to dense tensor.

View on GitHub (pinned to e006f5f0d5)