Regularization

[MaximilianPi]

Changes:

• importance function:
  old: rowSums( cov2cor(covariance)^2 ) / sqrt(ncol(covariance ))
  new: abs( rowSums( covariance ) - diag(covariance ) ) / ncol(covariance)
  • absolute: negative values don't work with the 0-1scaling
  • /ncol(sp): biotic part would scale still unfavorably with the number of species
  • using only covariances because we are interested in the biotic contributions
• Anova/sjSDM_cv/sjSDM, default sampling was increased to 1000. In the tuning function we are comparing a lot of different models with only small differences. Without a sufficient number of samples for the MC approximation, the model differences are lost in the MC approximation variations.
• regularization on the inverse covariance matrix works now properly