special_priors
========================

.. py:module:: jaxns.framework.special_priors

.. rubric:: :code:`jaxns.framework.special_priors`


.. rubric:: Module Contents

.. py:class:: Bernoulli(*, logits=None, probs=None, name = None)

   Bases: :py:obj:`SpecialPrior`


   The base prior class with public methods.


   .. py:attribute:: dist


.. py:class:: Beta(*, concentration0=None, concentration1=None, name = None)

   Bases: :py:obj:`SpecialPrior`


   The base prior class with public methods.


.. py:class:: Categorical(parametrisation, *, logits=None, probs=None, name = None)

   Bases: :py:obj:`SpecialPrior`


   The base prior class with public methods.

   Initialised Categorical special prior.

   :param parametrisation: 'cdf' is good for discrete params with correlation between neighbouring categories,
                           otherwise gumbel is better.
   :param logits: log-prob of each category
   :param probs: prob of each category
   :param name: optional name


   .. py:attribute:: dist


.. py:class:: ForcedIdentifiability(*, n, low=None, high=None, fix_left = False, fix_right = False, name = None)

   Bases: :py:obj:`SpecialPrior`


   Prior for a sequence of `n` random variables uniformly distributed on U[low, high] such that U[i,...] <= U[i+1,...].
   For broadcasting the resulting random variable is sorted on the first dimension elementwise.

   :param n: number of samples within [low,high]
   :param low: minimum of distribution
   :param high: maximum of distribution
   :param fix_left: if True, the leftmost value is fixed to `low`
   :param fix_right: if True, the rightmost value is fixed to `high`


   .. py:attribute:: n


   .. py:attribute:: low


   .. py:attribute:: high


   .. py:attribute:: fix_left
      :value: False



   .. py:attribute:: fix_right
      :value: False



.. py:class:: Poisson(*, rate=None, log_rate=None, name = None)

   Bases: :py:obj:`SpecialPrior`


   The base prior class with public methods.


   .. py:attribute:: dist


.. py:class:: UnnormalisedDirichlet(*, concentration, name = None)

   Bases: :py:obj:`SpecialPrior`


   Represents an unnormalised dirichlet distribution of K classes.
   That is, the output is related to the K-simplex via normalisation.

   X ~ UnnormalisedDirichlet(alpha)
   Y = X / sum(X) ==> Y ~ Dirichlet(alpha)


.. py:class:: Empirical(*, samples, support_min = None, support_max = None, resolution = 100, name = None)

   Bases: :py:obj:`SpecialPrior`


   Represents the empirical distribution of a set of 1D samples, with arbitrary batch dimension.


.. py:class:: TruncationWrapper(prior, low, high, name = None)

   Bases: :py:obj:`SpecialPrior`


   Wraps another prior to make it truncated.

   For truncated distribution the quantile transforms to:

       Q_truncated(p) = Q_untruncated( p * (F_truncated(high) - F_truncated(low)) + F_truncated(low))

   And the CDF transforms to:

       F_truncated(x) = (F_untruncated(x) - F_untruncated(low)) / (F_untruncated(high) - F_untruncated(low))


   .. py:attribute:: prior


   .. py:attribute:: low


   .. py:attribute:: high


   .. py:attribute:: cdf_low


   .. py:attribute:: cdf_diff


.. py:class:: ExplicitDensityPrior(*, axes, density, regular_grid = False, name = None)

   Bases: :py:obj:`SpecialPrior`


   The base prior class with public methods.