tensorflow_model_analysis/evaluators/query_metrics/ndcg.py [76:112]:
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
  def _calculate_dcg_at_k(self, k: int, sorted_values: List[float]) -> float:
    """Calculate the value of DCG@k.

    Args:
      k: The last position to consider.
      sorted_values: A list of gain values assumed to be sorted in the desired
        ranking order.

    Returns:
      The value of DCG@k.
    """
    return np.sum(
        np.array(sorted_values)[:k] / np.log2(np.array(range(2, k + 2))))

  def _calculate_ndcg(self, values: List[Tuple[int, float]], k: int) -> float:
    """Calculate nDCG@k, based on given rank and gain values.

    Args:
      values: A list of tuples representing rank order and gain values.
      k: The maximum position to consider in calculating nDCG

    Returns:
      The value of nDCG@k, for the given list of values.
    """
    max_rank = min(k, len(values))
    ranked_values = [
        gain for _, gain in sorted(values, key=lambda x: x[0], reverse=False)
    ]
    optimal_values = [
        gain for _, gain in sorted(values, key=lambda x: x[1], reverse=True)
    ]
    dcg = self._calculate_dcg_at_k(max_rank, ranked_values)
    optimal_dcg = self._calculate_dcg_at_k(max_rank, optimal_values)
    if optimal_dcg > 0:
      return dcg / optimal_dcg
    else:
      return 0
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -



tensorflow_model_analysis/metrics/ndcg.py [189:225]:
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
  def _calculate_dcg_at_k(self, k: int, sorted_values: List[float]) -> float:
    """Calculate the value of DCG@k.

    Args:
      k: The last position to consider.
      sorted_values: A list of gain values assumed to be sorted in the desired
        ranking order.

    Returns:
      The value of DCG@k.
    """
    return np.sum(
        np.array(sorted_values)[:k] / np.log2(np.array(range(2, k + 2))))

  def _calculate_ndcg(self, values: List[Tuple[int, float]], k: int) -> float:
    """Calculate NDCG@k, based on given rank and gain values.

    Args:
      values: A list of tuples representing rank order and gain values.
      k: The maximum position to consider in calculating nDCG

    Returns:
      The value of NDCG@k, for the given list of values.
    """
    max_rank = min(k, len(values))
    ranked_values = [
        gain for _, gain in sorted(values, key=lambda x: x[0], reverse=False)
    ]
    optimal_values = [
        gain for _, gain in sorted(values, key=lambda x: x[1], reverse=True)
    ]
    dcg = self._calculate_dcg_at_k(max_rank, ranked_values)
    optimal_dcg = self._calculate_dcg_at_k(max_rank, optimal_values)
    if optimal_dcg > 0:
      return dcg / optimal_dcg
    else:
      return 0
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -