EzTao

EzTao is a Python toolkit for conducting time-series analysis using continuous-time autoregressive moving average (CARMA) processes. It uses celerite (a fast gaussian processes regression library) to compute the likelihood of a set of proposed CARMA parameters given the input time series. Comparing to existing tools for performing CARMA analysis in Python which use Kalman filter to evaluate the likelihood function (e.g., Kali), EzTao offers a more scalable solution (see the celerite paper for a comparison).

EzTao consist of tools to both simulate CARMA processes and fit (maximum likelihood estimation or MLE) time series to CARMA models. The current version of EzTao is built on top of celerite, future versions will take advantage of celerite2 (still under active development) for a better integration with other probabilistic programing libraries such as PyMC3.

Installation

EzTao can be installed with pip using:

pip install eztao

Getting Started

Tutorials

Python API

Changelog

0.4.3 (2023-12-15)

  • Drop support for Python 3.7

  • Bump numba requirement to >=0.57.0.

  • New Features: Added seed options to gpSimRand, gpSimFull, and addNoise

  • Bug fixes: #74, #75

0.4.1 (2023-06-12)

  • Update reference to numpy bool/complex (#71)

  • Bug fixes: #50, #54, #59

0.4.0 (2021-07-19)

  • Fitting functions (i.e., drw_fit) are now fully modular (allow user provided optimization function, prior function and etc.)

  • A new addNoise function to simulated random noise given measurement errors.

  • Bug fixes: #44

  • API changes: n_iter -> n_opt in fitting functions.

0.3.0 (2021-01-07)

  • update parameter initialization in fit functions; removed de option #26, #27

  • add few utils functions #30, #25

  • add mcmc module #29

  • ts simulation now support linear error

  • added online documentation

0.2.3 (2020-12-08)

  • add methods to CARMA_term conversion between CARMA and poly space

  • fixed bugs and add tests for model 2nd order stat functions

  • close #2, close #10

0.2.1 (2020-12-05)

  • A bunch bug fixes in the ts.carma module

  • Improved _min_opt optimizer, now added to all fitting functions

  • Now using minimizer instead of differential evolution may result in more robust parameter estimates.

0.2.0 (2020-12-03)

Fixed some bugs and added new features.

  • Fixed the instability issue when fitting time series to models higher than DHO/CARMA(2,1)

  • Cleaned up the plotting module

  • Added PSD, ACVF, and SF functions

0.1.0 (2020-11-09)

First release!

  • Fully working CARMA kernels: DRW_term, DHO_term and CARMA_term

  • Functions to simulate CARMA time series given a kernel

  • Functions to fit arbitrary time series to CARMA models (still having instability issues with CARMA models higher than DHO/CARMA(2,1))