iEat: Instance-based machine learning method to predict biotic...

In david-beauchesne/iEat: Functions to apply a machine learning algorithm to predict species trophic interactions

Instance-based machine learning method to predict biotic interactions

Description

This method was published in Beauchesne et al. (2016) Life & Environment 66(3-4):333-342. The steps of the algorithm are visually presented in Figure 1 of the paper.

Usage

iEat(
  S0,
  S1,
  S2 = S1,
  sourceSim,
  targetSim = sourceSim,
  K = 5,
  minSim = 0.3,
  minWt = 0.5,
  predict = "full algorithm"
)

Arguments

S0	Matrix, catalogue of empirical data used to infer predictions
S1	Vector of taxa forming networking for which topology is predicted
S2	Vector of taxa in S1 for which we wish to predict resources (if unspecified, S2 == S1 and the whole network is predicted)
sourceSim	Matrix (numeric), source similarity matrix between S1 taxa and the union of S0 and S1 taxa, structure sourceSim[unique(S0,S1), S1]
targetSim	Matrix (numeric), source similarity matrix between S1 taxa and the union of S0 and S1 taxa, structure sourceSim[unique(S0,S1), S1] (if unspecified, targetSim == sourceSim and no similarity distinction between resources and consumers)
K	Integer, how many neighbours for K nearest neighbour evaluation
minSim	Integer, minimum similarity value accepted to consider taxa as similar (implemented to avoid unrealistic interactions)
minWt	Integer, minimum weight for candidate source to become a predicted source
predict	String, specifies whether the predictions are made from the "full algorithm", the "catalogue" or the "predictive" contribution. If unspecified, predict == 'full algorithm'. See Beauchesne et al. (2016) for more details. If predict == 'catalogue', the methodology corresponds to the approach presented by Gray et al. (2015).

Value

A dataframe with source taxa for which target predictions are made, target infered from catalogue data (empirical) and target infered from KNN algorithm

Author(s)

David Beauchesne

david-beauchesne/iEat documentation built on Nov. 5, 2022, 4:43 p.m.