STooDs-tools / STooDs

STooDs is a framework to build and estimate probabilistic models for data varying in Space, Time or other Dimensions.

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

STooDs

DOI

STooDs is a framework to build and estimate probabilistic models for data varying in Space, Time or other Dimensions. A typical STooDs case study can be described as follows:

  1. The target dataset contains the values taken by one of several predictand variables varying along one or more dimensions.
  2. Each variable in the dataset is assumed to follow a distribution whose parameters may vary along the same dimensions as the dataset.
  3. This variability is specified by regression formulas that may use parameters, covariates and processes.

Key properties of STooDs include:

  1. Possibility to model several random variables.
  2. Flexible choice of the distribution for each variable.
  3. Identifiability constraints enabling the use of 'hidden covariates' models (aka 'latent variables' or 'latent factors' models).
  4. Handling of censored data.
  5. Models adapted to large datasets are available (Nearest Neighbours Gaussian Process).

Usage

The current code is a computational engine and mostly aims at generating MCMC samples from the posterior distribution associated with the model and the data. It is implemented as an executable file controlled by a set of configuration text files. The easiest usage is to call STooDs at the command line by typing ./STooDs -wk "path/to/workspace/" (Linux) or STooDs -wk "path/to/workspace/" (Windows), where the workspace is the folder containing all configuration files. Further analyses such as exploring MCMC samples, making predictions or plotting results can be performed with other tools such as the R package RSTooDs.

Getting STooDs executable

A recent STooDs executable for your system (Windows or Linux) can be downloaded here.

Alternatively, the executable can be recompiled from sources using the provided makefile or Code::Blocks project. Files from the following projects are needed for this purpose:

  1. BMSL
  2. BaM
  3. miniDMSL (available soon!)

Managing STooDs configuration files

The easiest way to create the configuration files is to use the R package RSTooDs, which provides an user-friendly interface to manipulate data, specify a probabilistic model and estimate it: see the vignettes provided with this package.

It is also possible to create these configuration files by hand, or by implementing a user interface in a language other than R. The folder examples contains typical configuration files for a few simple case studies. A more complete documentation can be found in folder doc.

About

STooDs is a framework to build and estimate probabilistic models for data varying in Space, Time or other Dimensions.

License:GNU General Public License v3.0


Languages

Language:Fortran 98.9%Language:Makefile 1.1%