JAX based lib for sampling statistical distributions.
Project description
JAXampler 🧪 - A JAX-based statistical sampling toolkit
JAXampler is a statistical sampling toolkit built on top of JAX. It provides a set of high-performance sampling algorithms for a wide range of statistical distributions. JAXampler is designed to be easy to use and integrate with existing JAX workflows. It is also designed to be extensible, allowing users to easily add new sampling algorithms and statistical distributions.
JAXampler is currently in the early stages of development and is not yet ready for production use. However, we welcome contributions from the community to help us improve the library. If you are interested in contributing, please refer to our contribution guidelines.
Features
- 🚀 High-Performance Sampling: Leverage the power of JAX for high-speed, accurate sampling.
- 🧩 Versatile Algorithms: A wide range of sampling methods to suit various applications.
- 🔗 Easy Integration: Seamlessly integrates with existing JAX workflows.
Install
You may install the latest released version of JAXampler through pip by doing
pip3 install --upgrade jaxampler
You may install the bleeding edge version by cloning this repo, or doing
pip3 install --upgrade git+https://github.com/Qazalbash/jaxampler
If you would like to take advantage of CUDA, you will additionally need to install a specific version of JAX by doing
pip install --upgrade "jax[cuda12_pip]" -f https://storage.googleapis.com/jax-releases/jax_cuda_releases.html
Requirements
JAXampler requires Python 3.10 or later. It also requires the following packages:
jax>=0.4.0
jaxlib>=0.4.0
matplotlib>=3.8.0
tfp-nightly
tqdm
The test suite is based on pytest
. To run the tests, one needs to install pytest and run pytest
at the root directory of this repo.
Samplers
To sample from a distribution, import the corresponding sampler from the jaxampler.sampler
module and call the sample
method with the required arguments.
from jaxampler.rvs import Beta, Normal
from jaxampler.sampler import AcceptRejectSampler
scale = 1.35
N = 100_000
target_rv = Normal(mu=0.5, sigma=0.2)
proposal_rv = Beta(alpha=2, beta=2)
ar_sampler = AcceptRejectSampler()
samples = ar_sampler.sample(
target_rv=target_rv,
proposal_rv=proposal_rv,
scale=scale,
N=N,
)
JAXampler currently supports the following samplers:
- Inverse Transform Sampler
- Accept-Rejection Sampler
- Adaptive Accept-Rejection Sampler
- Metropolis-Hastings Sampler
- Hamiltonian Monte Carlo Sampler
- Slice Sampler
- Gibbs Sampler
- Importance Sampler
Random Variables
To create a new random variable, import the corresponding type from the jaxampler.rvs
i.e. DiscreteRV
and ContinuousRV
for discrete and continuous random variables respectively. Then, instantiate the random variable with the required parameters and implement the necessary methods (logpdf, cdf, and ppf etc). JAXampler currently supports the following random variables:
Discrete Random Variables
- Bernoulli
- Binomial
- Geometric
- Poisson
- Rademacher
Continuous Random Variables
- Beta
- Boltzmann
- Cauchy
- Chi
- Exponential
- Gamma
- Gumbel
- Laplace
- Log Normal
- Logistic
- Multivariate Normal
- Normal
- Pareto
- Rayleigh
- Student t
- Triangular
- Truncated Normal
- Truncated Power Law
- Uniform
- Weibull
Citing Jaxampler
To cite this repository:
@software{jaxampler2023github,
author = {Meesum Qazalbash},
title = {{JAXampler}: A JAX-based statistical sampling toolkit},
url = {http://github.com/Qazalbash/jaxampler},
version = {0.0.4},
year = {2023}
}
Contributors
Made with contrib.rocks.
Project details
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distribution
Built Distribution
Hashes for jaxampler-0.0.4-py3-none-any.whl
Algorithm | Hash digest | |
---|---|---|
SHA256 | fa882e68efe523b4ea18e996e5bbe44e16c4e466c045e384e6bbc4b1ac599c2b |
|
MD5 | f849331dfab7b7b1e04aef9cb365dc87 |
|
BLAKE2b-256 | f87ae484e04c6013fffd67cee0c763ed613cce691eda21847025f739d2a525a2 |