dataeval.metrics.bias.coverage
==============================

.. py:function:: dataeval.metrics.bias.coverage(embeddings, radius_type = 'adaptive', k = 20, percent = 0.01)

   Class for evaluating :term:`coverage<Coverage>` and identifying images/samples that are in undercovered regions.

   :param embeddings: A dataset in an ArrayLike format.
                      Function expects the data to have 2 dimensions, N number of observations in a P-dimesionial space.
   :type embeddings: ArrayLike, shape - (N, P)
   :param radius_type: The function used to determine radius.
   :type radius_type: {"adaptive", "naive"}, default "adaptive"
   :param k: Number of observations required in order to be covered.
             [1] suggests that a minimum of 20-50 samples is necessary.
   :type k: int, default 20
   :param percent: Percent of observations to be considered uncovered. Only applies to adaptive radius.
   :type percent: float, default 0.01

   :returns: Array of uncovered indices, critical value radii, and the radius for coverage
   :rtype: CoverageOutput

   :raises ValueError: If length of :term:`embeddings<Embeddings>` is less than or equal to k
   :raises ValueError: If radius_type is unknown

   .. note:: Embeddings should be on the unit interval [0-1].

   .. rubric:: Example

   >>> results = coverage(embeddings)
   >>> results.indices
   array([447, 412,   8,  32,  63])
   >>> results.critical_value
   0.8459038956941765

   Reference
   ---------
   This implementation is based on https://dl.acm.org/doi/abs/10.1145/3448016.3457315.

   [1] Seymour Sudman. 1976. Applied sampling. Academic Press New York (1976).