ars-dev 0.1.0

A JAX/numpy implementation of Adaptive Rejection Sampling (ARS) introduced by Gilks and Wild in 1992

Ars-dev

Ars-dev is a Python package designed for adaptive rejection sampling(ARS). This implementation references Adaptive Rejection Sampling for Gibbs Sampling(Gilks et al. 1992).

Installation

First clone this project locally with git

git clone https://github.berkeley.edu/yoshi-chiu/ars-dev.git

Then you can install the package via pip. First move in the directory, then run pip.

cd ars-dev/
python -m pip install .

Quick Start

Here is an example on how to run the package.

from ars import ars
from scipy.stats import norm
import numpy as np

domain = (-np.inf, np.inf)
n_samples = 10000
samples = ars(pdf=norm.pdf, domain=domain, n_samples=n_samples, is_logpdf=False, verbose=True)

The function also accepts pdf functions built with JAX. In that case, jax automatic differentiation is used in the sampling

from ars import ars
import jax.numpy as jnp

target_pdf = lambda x: jnp.exp(-x**2 / 2)
domain = (-jnp.inf, jnp.inf)

samples = ars(pdf=target_pdf, domain=domain, n_samples=1000, is_logpdf=False, verbose=True)

A function test_ars is also provided to run tests on easy to compute pdfs

from ars import test_ars
from scipy.stats import norm
import numpy as np

domain = (-np.inf, np.inf)
n_samples = 10000
samples, obj, fig = test_ars(pdf=norm.pdf, domain=domain, n_samples=n_samples, verbose=False)

Running Unit tests

To run all the unit tests, first move in the project directory then run pytest

pytest

Documentation

Documentation can be found at https://github.berkeley.edu/yoshi-chiu/ars-dev/blob/main/docs/report.pdf

License

This project is licensed under the MIT License.

---

---

---

(To be deleted)

Setup (For us)

Run this to install the package in editable mode

python -m pip install -e .

Running the function

TODO

• JIT ? (TBD)
• write testing for all non-trivial functions (Jordan)
• Setup as a package (Hongyu)
• write a README on the easy setup and initial simple examples (Hongyu)
• Write documentation, especially the main function with some simple examples (Yoshi)
• write a report documet explaining the implementation approach (Jordan for a good part of it)
• Add tests explanation and examples in report (Yoshi)
• Add in results plots and examples in report (Jordan, Yoshi)
• Improve the update of the arrays to avoid recomputing everything (not faster)
• Update code for non-jax functions (Jordan)