sbijax.simulators

sbijax.simulators contains several simulator models from the SBI literature.

hyperboloid()	Hyperboloid model.
jansen_rit([summarize_data])	Stochastic Jansen-Rit neural mass model.
mixture_model_with_distractors()	Mixture model with distractors.
sir([population_size, binomial_count, ...])	SIR model.
slcp()	Simple likelihood complex posterior model.
solar_dynamo([summarize_data])	Solar dynamo model.
tree()	Tree model.
two_moons()

sbijax.simulators.hyperboloid()

Hyperboloid model.

Constructs prior, simulator, and likelihood functions.

Returns:

returns a tuple of three objects. The first is a tfd.JointDistributionNamed serving as a prior distribution. The second is a simulator function that can be used to generate data. The third is the likelihood function.

References

Forbes, Florence, et al., Summary statistics and discrepancy measures for approximate Bayesian computation via surrogate posteriors, 2022

sbijax.simulators.jansen_rit(summarize_data=False)

Stochastic Jansen-Rit neural mass model.

Constructs prior and simulator functions.

Parameters:

summarize_data – if true returns the data from the simulator in a summarized version of 5 values. Otherwise, returns the infection counts of the ODE.

Returns:

returns a tuple of three objects. The first is a tfd.JointDistributionNamed serving as a prior distribution. The second is a simulator function that can be used to generate data. The third is None (since the likelihood is intractable and to be consistent with other models).

References

Ableidinger, Marko, et al., A stochastic version of the Jansen and Rit neural mass model: Analysis and numerics, 2017

sbijax.simulators.mixture_model_with_distractors()

Mixture model with distractors.

Constructs prior, simulator, and likelihood functions.

Returns:

returns a tuple of three objects. The first is a tfd.JointDistributionNamed serving as a prior distribution. The second is a simulator function that can be used to generate data. The third is the likelihood function.

References

Albert, Carlo, et al., Simulated Annealing ABC with multiple summary statistics, 2025

sbijax.simulators.sir(population_size=1000000, binomial_count=1000, initial_conditions=(999999, 1, 0), t_end=160, summarize_data=False)

SIR model.

Construct prior, simulator, and likelihood functions.

Parameters:

• population_size – the size of the population for the SIR model

• binomial_count – the number of Bernoulli trials for the Binomial likelihood

• initial_conditions – tuple of three integers that should sum to population_size

• t_end – end time of the ODE

• summarize_data – if true returns the data from the simulator in a summarized version of 5 values. Otherwise returns the infection counts of the ODE.

Returns:

returns a tuple of three objects. The first is a tfd.JointDistributionNamed serving as a prior distribution. The second is a simulator function that can be used to generate data. The third is the likelihood function.

References

Lueckmann, Jan-Matthis, et al., “Benchmarking Simulation-Based Inference”, 2021.

sbijax.simulators.slcp()

Simple likelihood complex posterior model.

Constructs prior, simulator, and likelihood functions.

Returns:

returns a tuple of three objects. The first is a tfd.JointDistributionNamed serving as a prior distribution. The second is a simulator function that can be used to generate data. The third is the likelihood function.

References

Papamakarios, George, Sequential Neural Likelihood: Fast Likelihood-free Inference with Autoregressive Flows, 2019

sbijax.simulators.solar_dynamo(summarize_data=False)

Solar dynamo model.

Constructs prior and simulator functions

Returns:

returns a tuple of three objects. The first is a tfd.JointDistributionNamed serving as a prior distribution. The second is a simulator function that can be used to generate data. The third is None (since the likelihood is intractable and to be consistent with other models).

References

Albert, Carlo, et al., Learning summary statistics for Bayesian inference with autoencoders, 2022

sbijax.simulators.tree()

Tree model.

Constructs prior, simulator and likelihood functions.

Returns:

returns a tuple of three objects. The first is a tfd.JointDistributionNamed serving as a prior distribution. The second is a simulator function that can be used to generate data. The third is None (since the likelihood is intractable and to be consistent with other models).

References

Gloeckler, Manuel, et al., All-in-one simulation-based inference, 2025

sbijax.simulators.two_moons()