Software

天文学のデータ解析や統計的推論のために開発している公開ソフトウェアです。特に、JAXによる自動微分と勾配ベースの推論を利用した物理モデリングやベイズ解析のためのコードを開発しています。

I develop publicly available software for astronomical data analysis and statistical inference, with a particular focus on physical modeling and Bayesian analysis using automatic differentiation in JAX and gradient-based inference.

jnkepler

多惑星系のための微分可能なN体モデル

JAXで実装したN体積分を用いて、多惑星系のトランジット時刻変動(TTV)、視線速度(RV)、光度曲線をモデル化するPythonパッケージです。自動微分によってモデルの勾配を効率よく計算でき、最適化やHamiltonian Monte Carlo(HMC/NUTS)による高次元のパラメータ推定に利用できます。

Differentiable N-body modeling for multi-planet systems

A Python package for modeling transit timing variations (TTVs), radial velocities (RVs), and light curves of multi-planet systems using an N-body integrator implemented in JAX. Automatic differentiation provides efficient model gradients for optimization and high-dimensional parameter inference with Hamiltonian Monte Carlo (HMC/NUTS).

関連論文 / Related paper: Masuda et al. (2024) ↗

jaxstar

JAXとHamiltonian Monte Carloを用いた恒星パラメータ推定

恒星進化モデルの等年齢線と測光・距離などの観測を比較し、恒星の質量、年齢、金属量などを推定するPythonパッケージです。恒星進化モデルをJAX上で補間し、NumPyroを用いたHMC/NUTSによって事後分布を効率的に計算します。

Stellar parameter inference with JAX and Hamiltonian Monte Carlo

A Python package for estimating stellar masses, ages, metallicities, and other parameters by comparing stellar evolution isochrones with observations such as photometry and distances. It interpolates stellar evolution models in JAX and efficiently samples posterior distributions using HMC/NUTS in NumPyro.

関連論文 / Related paper: Masuda (2022) ↗

numpyro-inferutils

NumPyroを用いた統計的推論のためのユーティリティ

NumPyroで記述した確率モデルから、尤度や事前分布の計算、制約付き・制約なしパラメータ間の変換、Fisher情報行列やHessianの計算、MAP推定などを行うための汎用的なPythonパッケージです。天文学に限らず、NumPyroを用いたモデルの解析や診断に利用できるように開発しています。

Utilities for statistical inference with NumPyro

A general-purpose Python package for evaluating likelihoods and priors, transforming between constrained and unconstrained parameters, computing Fisher information matrices and Hessians, and performing maximum a posteriori (MAP) estimation from probabilistic models written in NumPyro. It is designed to support model analysis and diagnostics with NumPyro across applications beyond astronomy.

jaxspec

Under development

恒星の高分散スペクトルの微分可能モデリング

恒星の合成スペクトルのグリッドを補間し、視線速度、射影自転速度 (vsini)、macroturbulence、装置分解能などを考慮して観測スペクトルをモデル化するJAXベースのコードです。NumPyroと組み合わせて、恒星パラメータや速度場のベイズ推定に利用しています。

現在は研究用に開発中のコードで、十分なテストやドキュメントを備えた一般ユーザー向けパッケージにはなっていません。

Differentiable modeling of high-resolution stellar spectra

JAX-based code that interpolates grids of synthetic stellar spectra and models observed spectra accounting for radial velocity, projected rotational velocity (vsini), macroturbulence, and instrumental resolution. Combined with NumPyro, it is used for Bayesian inference of stellar parameters and velocity fields.

The code is currently under development for research use and is not yet a package with sufficient testing and documentation for general users.

関連論文 / Related paper: Tomoyoshi et al. (2024) ↗