Benchmark amortized simulation-based inference (SBI) algorithms.
Datasets:
- train/test: parameter-observation pairs from the prior-simulator
joint distribution :math:`p(\theta, x)=p(\theta)p(x|\theta)`.
- reference (optional): observations math:`x_ref` and corresponding
samples from the reference posterior :math:`p(\theta|x_ref)`
(if available).
Solvers: amortized SBI algorithms trained on the joint to approximate
the posterior :math:`p(\theta|x)` for any observation :math:`x`.
Metrics:
- expected negative log likelihood (NLL) on test (stopping criterion)
and train datasets.
- C2ST, EMD, MMD on reference dataset (optional).
Solvers (Click on a solver to hide it or double click to hide all the others)
fmpe_lampe[layers=3]
Flow matching posterior estimator (FMPE) [1,2].
The solver trains a regression network to approximate a vector field inducing a
time-continuous normalizing flow between the posterior distribution and a standard
Gaussian distribution.
Implemented with the :mod:`lampe` package.
fmpe_lampe[layers=5]
Flow matching posterior estimator (FMPE) [1,2].
The solver trains a regression network to approximate a vector field inducing a
time-continuous normalizing flow between the posterior distribution and a standard
Gaussian distribution.
Implemented with the :mod:`lampe` package.
npe_lampe[flow=maf,transforms=1]
Neural posterior estimation (NPE) [1,2].
The solver trains a parametric conditional distribution :math:`q_\phi(\theta | x)`
to approximate the posterior distribution :math:`p(\theta | x)` of parameters given
observations.
Implementated with the :mod:`lampe` package.
npe_lampe[flow=maf,transforms=3]
Neural posterior estimation (NPE) [1,2].
The solver trains a parametric conditional distribution :math:`q_\phi(\theta | x)`
to approximate the posterior distribution :math:`p(\theta | x)` of parameters given
observations.
Implementated with the :mod:`lampe` package.
npe_lampe[flow=maf,transforms=5]
Neural posterior estimation (NPE) [1,2].
The solver trains a parametric conditional distribution :math:`q_\phi(\theta | x)`
to approximate the posterior distribution :math:`p(\theta | x)` of parameters given
observations.
Implementated with the :mod:`lampe` package.
npe_lampe[flow=nsf,transforms=1]
Neural posterior estimation (NPE) [1,2].
The solver trains a parametric conditional distribution :math:`q_\phi(\theta | x)`
to approximate the posterior distribution :math:`p(\theta | x)` of parameters given
observations.
Implementated with the :mod:`lampe` package.
npe_lampe[flow=nsf,transforms=3]
Neural posterior estimation (NPE) [1,2].
The solver trains a parametric conditional distribution :math:`q_\phi(\theta | x)`
to approximate the posterior distribution :math:`p(\theta | x)` of parameters given
observations.
Implementated with the :mod:`lampe` package.
npe_lampe[flow=nsf,transforms=5]
Neural posterior estimation (NPE) [1,2].
The solver trains a parametric conditional distribution :math:`q_\phi(\theta | x)`
to approximate the posterior distribution :math:`p(\theta | x)` of parameters given
observations.
Implementated with the :mod:`lampe` package.
npe_sbi[flow=maf,transforms=1]
Neural posterior estimation (NPE).
The solver trains a parametric conditional distribution :math:`q_\phi(\theta | x)`
to approximate the posterior distribution :math:`p(\theta | x)` of parameters given
observations.
Implementated with the :mod:`sbi` package.
npe_sbi[flow=maf,transforms=3]
Neural posterior estimation (NPE).
The solver trains a parametric conditional distribution :math:`q_\phi(\theta | x)`
to approximate the posterior distribution :math:`p(\theta | x)` of parameters given
observations.
Implementated with the :mod:`sbi` package.
npe_sbi[flow=maf,transforms=5]
Neural posterior estimation (NPE).
The solver trains a parametric conditional distribution :math:`q_\phi(\theta | x)`
to approximate the posterior distribution :math:`p(\theta | x)` of parameters given
observations.
Implementated with the :mod:`sbi` package.
npe_sbi[flow=nsf,transforms=1]
Neural posterior estimation (NPE).
The solver trains a parametric conditional distribution :math:`q_\phi(\theta | x)`
to approximate the posterior distribution :math:`p(\theta | x)` of parameters given
observations.
Implementated with the :mod:`sbi` package.
npe_sbi[flow=nsf,transforms=3]
Neural posterior estimation (NPE).
The solver trains a parametric conditional distribution :math:`q_\phi(\theta | x)`
to approximate the posterior distribution :math:`p(\theta | x)` of parameters given
observations.
Implementated with the :mod:`sbi` package.
npe_sbi[flow=nsf,transforms=5]
Neural posterior estimation (NPE).
The solver trains a parametric conditional distribution :math:`q_\phi(\theta | x)`
to approximate the posterior distribution :math:`p(\theta | x)` of parameters given
observations.
Implementated with the :mod:`sbi` package.
nre_lampe[layers=3]
Neural ratio estimation (NRE).
The solver trains a classifier to discriminate between pairs sampled from the joint
distribution :math:`p(\theta, x)` and the product of marginals :math:`p(\theta)
p(x)`.
Implementated with the :mod:`lampe` package.
nre_lampe[layers=5]
Neural ratio estimation (NRE).
The solver trains a classifier to discriminate between pairs sampled from the joint
distribution :math:`p(\theta, x)` and the product of marginals :math:`p(\theta)
p(x)`.
Implementated with the :mod:`lampe` package.