This release is a pre-release and may not be stable for production use.
AeMCMC is a Python library that automates the construction of samplers for Aesara graphs that represent statistical models.
Features
This project is currently in an alpha state, but the basic features/objectives are currently as follows:
Provide utilities that simplify the process of constructing Aesara graphs/functions for posterior and posterior predictive sampling
Host a wide array of “exact” posterior sampling steps (e.g. Gibbs steps, scale-mixture/decomposition-based conditional samplers, etc.)
Build a framework for identifying and composing said sampler steps and enumerating the possible samplers for an arbitrary model
Overall, we would like this project to serve as a hub for community-sourced specialized samplers and facilitate their general use.
Getting started
Using AeMCMC, one can construct sampling steps from a graph containing Aesara RandomVariables. AeMCMC analyzes the model graph and possibly rewrites it to find the most suitable sampler.
AeMCMC can recognize closed-form posteriors; for instance the following Beta-Binomial model amounts to sampling from a Beta distribution:
import aesara
import aemcmc
import aesara.tensor as at
srng = at.random.RandomStream(0)
p_rv = srng.beta(1., 1., name="p")
Y_rv = srng.binomial(10, p_rv, name="Y")
y_vv = Y_rv.clone()
y_vv.name = "y"
sampler, initial_values = aemcmc.construct_sampler({Y_rv: y_vv}, srng)
p_posterior_step = sampler.sample_steps[p_rv]
aesara.dprint(p_posterior_step)
# beta_rv{0, (0, 0), floatX, False}.1 [id A]
# |RandomGeneratorSharedVariable(<Generator(PCG64) at 0x7F77B2831200>) [id B]
# |TensorConstant{[]} [id C]
# |TensorConstant{11} [id D]
# |Elemwise{add,no_inplace} [id E]
# | |TensorConstant{1.0} [id F]
# | |y [id G]
# |Elemwise{sub,no_inplace} [id H]
# |Elemwise{add,no_inplace} [id I]
# | |TensorConstant{1.0} [id F]
# | |TensorConstant{10} [id J]
# |y [id G]
sample_fn = aesara.function([y_vv], p_posterior_step)
AeMCMC also contains a database of Gibbs samplers that can be used to sample some models more efficiently than a general-purpose sampler like NUTS would:
import aemcmc
import aesara
import aesara.tensor as at
srng = at.random.RandomStream(0)
X = at.matrix("X")
# Horseshoe prior for `beta_rv`
tau_rv = srng.halfcauchy(0, 1, name="tau")
lmbda_rv = srng.halfcauchy(0, 1, size=X.shape[1], name="lambda")
beta_rv = srng.normal(0, lmbda_rv * tau_rv, size=X.shape[1], name="beta")
a = at.scalar("a")
b = at.scalar("b")
h_rv = srng.gamma(a, b, name="h")
# Negative-binomial regression
eta = X @ beta_rv
p = at.sigmoid(-eta)
Y_rv = srng.nbinom(h_rv, p, name="Y")
y_vv = Y_rv.clone()
y_vv.name = "y"
sampler, initial_values = aemcmc.construct_sampler({Y_rv: y_vv}, srng)
# `sampler.sample_steps` contains the sample step for each random variable
print(sampler.sample_steps[h_rv])
# h_posterior
# `sampler.stages` contains the sampling kernels sorted by scan order
print(sampler.stages)
# {HorseshoeGibbsKernel: [tau, lambda], NBRegressionGibbsKernel: [beta], DispersionGibbsKernel: [h]}
# Build a function that returns new samples
to_sample_rvs = [tau_rv, lmbda_rv, beta_rv, h_rv]
inputs = [a, b, X, y_vv] + [initial_values[rv] for rv in to_sample_rvs]
outputs = [sampler.sample_steps[rv] for rv in to_sample_rvs]
sample_fn = aesara.function(inputs, outputs, updates=sampler.updates)
In case no specialized sampler is found, AeMCMC assigns the NUTS sampler to the remaining variables. AeMCMC reparametrizes the model automatically to improve sampling if needed:
import aemcmc
import aesara
import aesara.tensor as at
srng = at.random.RandomStream(0)
mu_rv = srng.normal(0, 1, name="mu")
sigma_rv = srng.halfnormal(0.0, 1.0, name="sigma")
Y_rv = srng.normal(mu_rv, sigma_rv, name="Y")
y_vv = Y_rv.clone()
sampler, initial_values = aemcmc.construct_sampler({Y_rv: y_vv}, srng)
print(sampler.sample_steps.keys())
# dict_keys([sigma, mu])
print(sampler.stages)
# {NUTSKernel: [sigma, mu]}
print(sampler.parameters)
# {NUTSKernel: (step_size, inverse_mass_matrix)}
# Build a function that returns new samples
step_size, inverse_mass_matrix = list(sampler.parameters.values())[0]
inputs = [
initial_values[mu_rv],
initial_values[sigma_rv],
y_vv,
step_size,
inverse_mass_matrix
]
outputs = [sampler.sample_steps[mu_rv], sampler.sample_steps[sigma_rv]]
sample_fn = aesara.function(inputs, outputs, updates=sampler.updates)
Installation
The latest release of AeMCMC can be installed from PyPI using pip:
pip install aemcmc
Or via conda-forge:
conda install -c conda-forge aemcmc
The current development branch of AeMCMC can be installed from GitHub, also using pip:
pip install git+https://github.com/aesara-devs/aemcmc
Metadata
Release files for aemcmc-nightly 0.0.6.dev20221211
For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.
Source distribution (sdist)
| File | Size | Uploaded | Provenance | Actions |
|---|---|---|---|---|
| aemcmc-nightly-0.0.6.dev20221211.tar.gz | 43.4 kB | Without provenance | Details for aemcmc-nightly-0.0.6.dev20221211.tar.gz |
Release files / aemcmc-nightly-0.0.6.dev20221211.tar.gz
| Download URL | aemcmc-nightly-0.0.6.dev20221211.tar.gz |
|---|---|
| Size | 43.4 kB |
| Tags | Source |
|
SHA-256 checksum How to use checksums |
b61f8ae2158f57888cc578ae595f7a001d17bb6dd0d089c1b2183122a8a4128e
|
|
BLAKE2b-256 checksum How to use checksums |
d396969bc994cbc3e58a671001ab5c37373f979cdde30ded3e7ab357879d000b
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
twine/4.0.2 CPython/3.9.16
|
Security for aemcmc-nightly 0.0.6.dev20221211
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
No provenance found
None of the files in this release were uploaded with provenance metadata.
A note on package security
PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.
To learn more about protecting yourself from malicious actors, visit our package security documentation.