pyepo.func.jax.cave
===================

.. py:module:: pyepo.func.jax.cave

.. autoapi-nested-parse::

   Cone-aligned vector estimation (CaVE) loss for binary linear programs



Attributes
----------

.. autoapisummary::

   pyepo.func.jax.cave.CaVE


Classes
-------

.. autoapisummary::

   pyepo.func.jax.cave.coneAlignedCosine


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

.. py:class:: coneAlignedCosine(optmodel, max_iter=3, solve_ratio=1.0, inner_ratio=0.2, processes=1, reduction: pyepo.func.runtime.Reduction = 'mean')

   Bases: :py:obj:`pyepo.func.jax.abcmodule.optModule`


   Cone-Aligned Vector Estimation (CaVE) loss for binary linear programs.

   Projects the sense-flipped predicted cost onto the polyhedral cone spanned
   by the binding-constraint normals at the true optimal vertex (a Clarabel
   QP) and minimizes :math:`1 - \cos(-\hat{\mathbf{c}}, \mathrm{proj})`.
   The projection is detached, so the gradient flows only through the cosine.

   Reference: Tang & Khalil (2024)
   `<https://link.springer.com/chapter/10.1007/978-3-031-60599-4_12>`_


   .. py:attribute:: max_iter
      :value: 3



   .. py:attribute:: solve_ratio


   .. py:attribute:: inner_ratio


   .. py:method:: forward(pred_cost, tight_ctrs)

      Forward pass



.. py:data:: CaVE