Discrimination/Classification in very high dimension with linear and quadratic rules.

This package provides an implementation of Linear discriminant analysis and quadratic discriminant analysis that works fine in very high dimension (when there are many more variables than observations).

Optimisation with Continuous Convex Piecewise (Linear and Quadratic) Functions

Continuous convex piecewise linear (ccpl) resp. quadratic (ccpq) functions can be implemented with sorted breakpoints and slopes. This includes functions that are ccpl (resp. ccpq) on a convex set (i.e. an interval or a point) and infinite out of the domain. These functions can be very useful for a large class of optimisation problems. Efficient manipulation (such as log(N) insertion) of such data structure is obtained with map standard template library of C++ (that hides balanced trees). This package is a wrapper on such a class based on Rcpp modules.

Poisson Lognormal Models

The Poisson-lognormal model and variants (Chiquet, Mariadassou and Robin, 2021 ) can be used for a variety of multivariate problems when count data are at play, including principal component analysis for count data, discriminant analysis, model-based clustering and network inference. Implements variational algorithms to fit such models accompanied with a set of functions for visualization and diagnostic.

Three-Dimensional Exploratory Projection Pursuit

Exploratory projection pursuit is a method to discovers structure in multivariate data. At heart this package uses a projection index to evaluate how interesting a specific three-dimensional projection of multivariate data (with more than three dimensions) is. Typically, the main structure finding algorithm starts at a random projection and then iteratively changes the projection direction to move to a more interesting one. In other words, the projection index is maximised over the projection direction to find the most interesting projection. This maximum is, though, a local maximum. So, this code has the ability to restart the algorithm from many different starting positions automatically. Routines exist to plot a density estimate of projection indices over the runs, this enables the user to obtain an idea of the distribution of the projection indices, and, hence, which ones might be interesting. Individual projection solutions, including those identified as interesting, can be extracted and plotted individually. The package can make use of the mclapply() function to execute multiple runs in parallel to speed up index discovery. Projection pursuit is similar to independent component analysis. This package uses a projection index that maximises an entropy measure to look for projections that exhibit non-normality, and operates on sphered data. Hence, information from this package is different from that obtained from principal components analysis, but the rationale behind both methods is similar. Nason, G. P. (1995) .

Data Sonification - Turning Data into Sound

Sonification (or audification) is the process of representing data by sounds in the audible range. This package provides the R function sonify() that transforms univariate data, sampled at regular or irregular intervals, into a continuous sound with time-varying frequency. The ups and downs in frequency represent the ups and downs in the data. Sonify provides a substitute for R's plot function to simplify data analysis for the visually impaired.

Robust Mixture Model

Algorithms for estimating robustly the parameters of a Gaussian, Student, or Laplace Mixture Model.

Continued Fractions

Various utilities for evaluating continued fractions.

Bayesian Analysis of Computer Code Output (BACCO)

The BACCO bundle of packages is replaced by the BACCO package, which provides a vignette that illustrates the constituent packages (emulator, approximator, calibrator) in use.

The Davies Quantile Function

Various utilities for the Davies distribution.

Bayesian Prediction of Complex Computer Codes

Performs Bayesian prediction of complex computer codes when fast approximations are available. It uses a hierarchical version of the Gaussian process, originally proposed by Kennedy and O'Hagan (2000), Biometrika 87(1):1.