jpt.learning.pruning.xi
=======================

.. py:module:: jpt.learning.pruning.xi

.. autoapi-nested-parse::

   Xi-correlation-based pruning criterion.

   Uses Chatterjee's xi coefficient to decide
   whether a split is justified by statistically
   significant functional dependence between
   features and targets in the current partition.



Classes
-------

.. autoapisummary::

   jpt.learning.pruning.xi.XiPruningCriterion


Module Contents
---------------

.. py:class:: XiPruningCriterion(alpha: float = 0.05, min_n: int = 30)

   Prune-or-split callback based on xi.

   Tests whether any feature-target pair shows
   significant functional dependence in the
   current partition. Under H0 (independence),
   sqrt(n) * xi ~ N(0, 2/5).

   :param alpha: significance level
   :param min_n: minimum samples to apply test


   .. py:attribute:: z_crit
      :type:  float


   .. py:attribute:: min_n
      :type:  int
      :value: 30



   .. py:method:: __call__(jpt: JPT, partition: JPTPartition, indices: numpy.ndarray, data: numpy.ndarray) -> bool

      Return True to prune (stop splitting).

      :param jpt:       the JPT being learned
      :param partition: current data partition
      :param indices:   index buffer mapping
                        positions to data rows
      :param data:      the full training data
                        array (n_samples x n_vars)
      :returns:         True if the node should
                        become a leaf