anglican.ipmcmc
Interacting particle Markov chain Monte Carlo Options: :number-of-particles (2 by default) - Number of particles per sweep :number-of-nodes (32 by default) - Number of nodes running SMC and CSMC. :number-of-csmc-nodes (nil by default) - Number of nodes running as CSMC. Must be between 1 and (- :number-of-nodes 1). Defaults to (/ :number-of-nodes 2) when not specified. :all-particles? (true by default) - Return all particles, instead of 1 particle per sweep on each node. Note that even when :all-particles? is false, particles are still weighted due to Rao- Blackwellization of the Gibbs updates for the CSMC indices. :pool (:builtin by default) - Threadpool argument for pmap operation over nodes. Defaults to creating a pool of containing (+ (ncpus) 2) threads. See com.climate.claypoole/pmap for further info.
Interacting particle Markov chain Monte Carlo
Options:
:number-of-particles (2 by default)
- Number of particles per sweep
:number-of-nodes (32 by default)
- Number of nodes running SMC and CSMC.
:number-of-csmc-nodes (nil by default)
- Number of nodes running as CSMC. Must be between
1 and (- :number-of-nodes 1). Defaults to
(/ :number-of-nodes 2) when not specified.
:all-particles? (true by default)
- Return all particles, instead of 1 particle per sweep
on each node. Note that even when :all-particles? is
false, particles are still weighted due to Rao-
Blackwellization of the Gibbs updates for the CSMC indices.
:pool (:builtin by default)
- Threadpool argument for pmap operation over nodes.
Defaults to creating a pool of containing (+ (ncpus) 2)
threads. See com.climate.claypoole/pmap for further info.gibbs-update-csmc-indicesclj
(gibbs-update-csmc-indices log-Zs number-of-csmc-nodes)Performs a Gibbs sweep on the indices of conditional nodes by sampling each index conditioned on the values of the other indices.
Returns a pair [csmc-indices zeta-sums] in which the csmc-indices is a vector of indices for newly selected conditional nodes and zeta-sums is a vector of weights for each node (needed when returning all particles).
Performs a Gibbs sweep on the indices of conditional nodes by sampling each index conditioned on the values of the other indices. Returns a pair [csmc-indices zeta-sums] in which the csmc-indices is a vector of indices for newly selected conditional nodes and zeta-sums is a vector of weights for each node (needed when returning all particles).
norm-expclj
(norm-exp log-weights)Normalized exponential. Accepts a collection of log weights. Returns a pair [ps log-Z] in which ps is a sequence of normalized probabilities and log-Z is the log mean weight. If all weights are -infinity then they are assumed to be equal.
Normalized exponential. Accepts a collection of log weights. Returns a pair [ps log-Z] in which ps is a sequence of normalized probabilities and log-Z is the log mean weight. If all weights are -infinity then they are assumed to be equal.
sweepclj
(sweep algorithm prog value number-of-particles all-particles? retained-state)Performs a sequential Monte Carlo or conditional sequential Monte Carlo sweep. Returns a pair [results log-Z] in which results is a sequence of result records and log-Z is an estimate of the log marginal likelihood.
Performs a sequential Monte Carlo or conditional sequential Monte Carlo sweep. Returns a pair [results log-Z] in which results is a sequence of result records and log-Z is an estimate of the log marginal likelihood.
cljdoc is a website building & hosting documentation for Clojure/Script libraries
× close