jexplore.steps.colored_alt

jexplore.steps.colored_alt#

This module defines an alternate version of jexplore.steps.colored.ColoredSC and of its child classes jexplore.steps.stretch.Stretch and jexplore.steps.de.DEStep. The main difference is that the proposed partners for each chain are distinct.

Classes#

ColoredAlt

Class implementing a MH steps based on stretch proposal

StretchAlt

Class implementing a MH steps based on stretch proposal. This is an

DEStepAlt

Class implementing a Differential evolution step. This is an

Module Contents#

class ColoredAlt[Tepoch: jexplore.sampling.EpochMH, Tstate: jexplore.sampling.StateMH, Tsampling: jexplore.sampling.SamplingMH](ngroups=2, permute=False, duplicate=False, adjust=False)[source]#

Bases: jexplore.steps.colored.ColoredSC[Tepoch, Tstate, Tsampling]

Class implementing a MH steps based on stretch proposal

Parameters:

  • a – stretch proposal a parameter

  • ngroups (int) – number of groups. Default 2.

  • permute (bool) – if true walkers are permuted at each iteration.

  • duplicate (bool)

  • adjust (bool)

get_partners(key, state, group, cgroup)[source]#

Method for getting partners samples fore each chain of a group. This alternate implementation gets the first self.npart * group.size // ntemps chains out of the cgroup chains for each temperatures, after permuting them and uses them as partners of the group chains of the same temperature. With respect to the base implementation of this method, this one always returns distinct partners for each group. For this reason ngroups must be larger than self.npart + 1 (this has to be implemented by the children classes).

Parameters:

  • key (jax.Array) – PRNG key

  • state (Tstate) – current state

  • group (jax.Array) – group chains

  • cgroup (jax.Array) – complementary group chains

Returns:

the parners as an array with shape (self.npars, group.size, dim)

Return type:

jax.Array

abstract proposal(key, state, group, cgroup)[source]#

Proposal restricted to chains of one color. This is just a prototype.

Parameters:

  • key (jax.Array) – PRNG key

  • state (Tstate) – current state

  • group (jax.Array) – current color group

  • cgroup (jax.Array) – complementary chains for each chain in the group.

Returns:

PNRG key, proposed state, transition probability for each chain.

Return type:

tuple[jax.Array, Tstate, jax.Array]

class StretchAlt[Tepoch: jexplore.sampling.EpochMH, Tstate: jexplore.sampling.StateMH, Tsampling: jexplore.sampling.SamplingMH](a=2.0, ngroups=2, permute=False)[source]#

Bases: jexplore.steps.stretch.Stretch[Tepoch, Tstate, Tsampling]

Class implementing a MH steps based on stretch proposal. This is an alternate version inwhich each chain of a color group receives a distinct partner from the complementary group chains.

Parameters:

  • a (float) – stretch proposal a parameter

  • ngroups (int) – number of groups. Default 2.

  • permute (bool) – if true walkers are permuted at each iteration.

get_partners#

Method for getting partners samples fore each chain of a group. This implementation simply draws randomly one partner for each chain in the group among those in the chains of complementary group with the same temperature.

Parameters:

  • key – PRNG key

  • state – current state

  • group – group chains

  • cgroup – complementary group chains

Returns:

the parners as an array with shape (self.npars, group.size, dim)

class DEStepAlt[Tepoch: jexplore.sampling.EpochMH, Tstate: jexplore.sampling.StateMH, Tsampling: jexplore.sampling.SamplingMH](gamma=2.38, ngroups=3, permute=False)[source]#

Bases: jexplore.steps.de.DEStep[Tepoch, Tstate, Tsampling]

Class implementing a Differential evolution step. This is an alternate version inwhich each chain of a color group receives a distinct couple of partner from the complementary group chains.

Parameters:

  • gamma (float) – \(\gamma\) scale parameter

  • ngroups (int) – number of groups. Default 2.

  • permute (bool) – if true walkers are permuted at each iteration.

get_partners#

Method for getting partners samples fore each chain of a group. This implementation simply draws randomly one partner for each chain in the group among those in the chains of complementary group with the same temperature.

Parameters:

  • key – PRNG key

  • state – current state

  • group – group chains

  • cgroup – complementary group chains

Returns:

the parners as an array with shape (self.npars, group.size, dim)