Ctrl+K

blackjax.smc.ess

blackjax.smc.ess#

All things related to SMC effective sample size

Module Contents#

Functions#

ess(→ float)

log_ess(→ float)

Compute the effective sample size.

ess_solver(logdensity_fn, particles, target_ess, ...)

ESS solver for computing the next increment of SMC tempering.
ess(log_weights: blackjax.types.Array) → float[source]#
log_ess(log_weights: blackjax.types.Array) → float[source]#

Compute the effective sample size.

Parameters:

log_weights (1D Array) – log-weights of the sample

Returns:

log_ess – The logarithm of the effective sample size

Return type:

float

ess_solver(logdensity_fn: Callable, particles: blackjax.types.ArrayLikeTree, target_ess: float, max_delta: float, root_solver: Callable)[source]#

ESS solver for computing the next increment of SMC tempering.

Parameters:

  • logdensity_fn (Callable) – The log probability function we wish to sample from.

  • particles (SMCState) – Current state of the tempered SMC algorithm

  • target_ess (float) – The relative ESS targeted for the next increment of SMC tempering

  • max_delta (float) – Max acceptable delta increment

  • root_solver (Callable, optional) – A solver to find the root of a function, takes a function f, a starting point delta0, a min value min_delta, and a max value max_delta. Default is BFGS minimization of f ** 2 and ignores min_delta and max_delta.

Returns:

delta – The increment that solves for the target ESS

Return type:

float

Contents