README.md

In jeffdaniel/ppmnet: Regularized Spatial Point Process Models

ppmnet

ppmnet is an R package for fitting regularized spatial point process models.

Models are fit via penalized likelihood for Poisson point processes, and via penalized composite likelihood for Gibbs point processes. The model-fitting procedure is carried out by glmnet, which implements the lasso and elastic net penalties. A number of methods are provided for plotting, prediction, validation, and model selection on the basis of (composite) information criteria.

This package closely conforms to the conventions of the spatstat package, and relies heavily on spatstat classes and functions. As such, use of ppmnet should be straightforward to those familiar with spatstat.

Installation

You can install ppmnet from GitHub with:

devtools::install_github("jeffdaniel/ppmnet")

Example

Fit a regularized inhomogeneous Poisson point process model to the bei dataset included with spatstat.

fit <- ppmnet(quadscheme(bei), bei.extra)

Select the optimal model from the regularization path on the basis of minimum AIC and plot the model’s predicted intensity surface.

lambda <- select(fit, "AIC")$lambda
#> Warning: minimum value of tuning parameter selected
plot(predict(fit, bei.extra, s = lambda))

Examine the smoothed residual field for the selected model.

plot(Smooth(residuals(fit, s = lambda)))

Citation

Daniel, J., Horrocks, J., and Umphrey, G. J. Penalized composite likelihoods for inhomogeneous Gibbs point process models. Computational Statistics and Data Analysis, 124, 104–116. https://doi.org/10.1016/j.csda.2018.02.005

jeffdaniel/ppmnet documentation built on Aug. 14, 2019, 6:31 a.m.