Build a sampler for your model

aemcmc.basic.construct_sampler(obs_rvs_to_values, srng)[source]

Eagerly construct a sampler for a given set of observed variables and their observations.

Parameters:

obs_rvs_to_values – A dict of variables that maps stochastic elements (e.g. RandomVariables) to symbolic Variables representing their observed values.

Returns:

  • A dict that maps each random variable to its sampler step and

  • any updates generated by the sampler steps.

The Sampler object

construct_sampler returns a Sampler object that contains the graphs for the variables’ sampling steps and the updates to pass to aesara.function:

class aemcmc.types.Sampler(sample_steps, updates=<factory>, parameters=<factory>)[source]

A class that tracks sampling steps and their parameters.

parameters: Dict[aemcmc.types.SamplingStep, Tuple[aesara.tensor.var.TensorVariable]]

Parameters needed by the sampling steps.

sample_steps: Dict[aesara.tensor.var.TensorVariable, aesara.tensor.var.TensorVariable]

A map between measures and their updated value under the current sampling scheme.

stages: Dict[aemcmc.types.SamplingStep, List[aesara.tensor.var.TensorVariable]]

A list of the sampling stages sorted in scan order.

updates: Optional[Dict[aesara.graph.basic.Variable, aesara.tensor.var.TensorVariable]]

Updates to be passed to aesara.function