## S3 method for class 'default':
performance(object, \dots)
## S3 method for class 'default':
crossval(object, \dots)wa, wapls etc.performance returns a list with a named matrix object which contains the following columns:crossval, the list returned by performance also contains a matrix named crossval, which contains the above statistics calculated for the cross-validation predictions.
Function crossval returns an object of the original class and adds the following named elements:rand.t.test is a generic function that performs a randomisation t-test to test the significance of a cross-validated model, after van der Voet (1994). Methods exist for WA, WAPLS and IKFA.rioja implements a number of numerical methods for inferring the value of an environmental variable from a set of sepecies abundances, given a modern training set of species data and associated environmental values. In palaeoecology these are known as "transfer functions" or "inference models" and are used to hindcast or "reconstruct" past environmental conditions from sub-fossil species assemblages preserved in sediment cores. The techniques included are weighted averaging (WA), partial least squares (PLS) and weighted average partial least squared (WAPLS), Imbrie and Kipp Factor Analysis (IKFA) a form of principal components regression, Maximum Likelihood response Curves (MLRC), and the Modern Analogue Technique (MAT, a form of k-NN non-parametric regression (see Juggins & Birks (2010) for a review).
The techniques are implemented in a consistent way and include functions for fitting a model to a training set of species and environmental data, with the function name named after the technique: that is, WA fits a weighted averaging model. Any model can be cross-validated using the crossval function, which allows internal cross-validation using leave-one-out, leave-n-out, bootstrapping or h-block cross-validation. There are a number of generic functions that can be used to summarise and diagnose the models: (print, summary, performance and plot. Some techniques have additional diagnostic functions such as screeplot and rand.t.test to help estimate the approproate number of components (WAPLS), factors (IKFA) or number of analogues (IKFA).
Predictions for new species data can be made using predict, with an option to calculate sample-specific errors using bootstrapping, after the method described in Birks et al. (1990).