mtd

Helper classes for Bayesian model-to-data comparison.


License
MIT
Install
pip install mtd==0.1.dev0

Documentation

mtd

Helper classes for Bayesian model-to-data comparison.

Based on Computer Model Calibration Using High-Dimensional Output (2008) and A Bayesian Approach for Parameter Estimation and Prediction using a Computationally Intensive Model (2014) by Dave Higdon et al.

WARNING

I made this for my personal use, in particular for my papers http://inspirehep.net/record/1342465 and http://inspirehep.net/record/1458287. Some example scripts using this library: https://github.com/jbernhard/mtd-paper/blob/master/mcmc/train-and-calibrate and https://github.com/jbernhard/qm2015/blob/master/calibration/calibrate.

While mtd certainly served its purpose, it lacks features and flexibility, and I don't plan to develop it further.

Features

  • Emulation of multivariate models via Gaussian process regression and principal component analysis.
  • Automated emulator training and calibration to experimental data.
  • Intuitive system for specifying priors.
  • Fully multithreaded.
  • Built on Numpy+Scipy and Dan Foreman-Mackey's excellent MCMC toolkit emcee and Gaussian process library george.