bayesbay.samplers.ParallelTempering

class bayesbay.samplers.ParallelTempering(temperature_max=5, chains_with_unit_temperature=0.4, swap_every=500)

High-level class to be used to sample the posterior by means of reversible-jump Markov chain Monte Carlo accelerated with parallel tempering.

See references below for details on parallel tempering [1], [2].

Parameters:

  • temperature_max (Number) – the maximum temperature attributed to the chains

  • chains_with_unit_temperature (float) – the fraction of chains having unit temperature, 0.4 by default (i.e. 40% of the chains)

  • swap_every (int) – the frequency with which the chain temperatures are randomly chosen and possibly swapped during the sampling, by default 500 iterations

References

Reference Details

chains

the MarkovChain instances of the current Bayesian inference

parallel_config

customized parallel configuration dictionary that will be passed to joblib.Parallel

add_on_begin_iteration(func)

adds a custom function that will be called at the beginning of each Markov chain iteration (see on_begin_iteration())

add_on_end_advance_chain(func)

adds a custom function that will be called at the end of the batch of Markov chain iterations taking place after calling advance_chain()

add_on_end_iteration(func)

adds a custom function that will be called at the end of each Markov chain iteration (see on_end_iteration())

add_on_initialize(func)

adds a custom function that will be called at the end of the chains initialization (see initialize())

Parameters:

func (Callable) – Function that takes as arguments a Sampler and a list of MarkovChains

advance_chain(n_iterations, burnin_iterations=0, save_every=100, verbose=True, print_every=100)

advances the Markov chains for a given number of iterations

Parameters:

  • n_iterations (int) – the number of iterations to advance

  • burnin_iterations (int, optional) – the iteration number from which we start to save samples, by default 0

  • save_every (int, optional) – the frequency in which we save the samples, by default 100

  • verbose (bool, optional) – whether to print the progress during sampling or not, by default True

  • print_every (int, optional) – the frequency with which we print the progress and information during the sampling, by default 100 iterations

  • parallel_config (dict, optional) – keyword arguments passed to joblib.Parallel. Ignored when len(self.chains) is 1

Returns:

the Markov chains

Return type:

List[BaseMarkovChain]

begin_iteration(chain)

calls on_begin_iteration() before beginning the current Markov chain iteration

Parameters:

chain (bayesbay.BaseMarkovChain) – Markov chain used to sample the posterior

end_advance_chain()

calls on_end_advance_chain() before concluding the batch of Markov chain iterations taking place after calling advance_chain()

end_iteration(chain)

calls on_end_iteration() before passing to the next Markov chain iteration.

Parameters:

chain (bayesbay.BaseMarkovChain) – Markov chain used to sample the posterior

initialize(chains)

initializes the given Markov Chains by calling on_initialize()

Parameters:

chains (List[BaseMarkovChain]) – List of Markov chains used to sample the posterior

on_begin_iteration(chain)

defines the behaviour of the sampler at the beginning of each Markov chain iteration

Parameters:

chain (BaseMarkovChain) – Markov chain used to sample the posterior

on_end_advance_chain()

defines the behaviour of the sampler at the end of the batch of Markov chain iterations taking place after calling advance_chain()

Parameters:

chain (BaseMarkovChain) – Markov chain used to sample the posterior

on_end_iteration(chain)

defines the behaviour of the sampler at the end of each Markov chain iteration

Parameters:

chain (BaseMarkovChain) – Markov chain used to sample the posterior

on_initialize(chains)

defines the behaviour of the sampler at the initialization of the Markov chains

Parameters:

chains (List[BaseMarkovChain]) – List of Markov chains used to sample the posterior

run(n_iterations, burnin_iterations=0, save_every=100, verbose=True, print_every=100)

function that allows for running the Bayesian inference by calling advance_chain().

Important

To work properly, a custom Sampler has to call within this function the method advance_chain().

back to top