NEWS.md
randomCoefPlot functionality of constrained and concurrent ordination models with random slopes. Currently not supported for models with quadratic responsescoef now renames parameter estimates with more intuitive names and allows to subset the parameter list with namesOrdination with predictors (num.RR,num.lv.c) is now implemented with constrained optimization routines (alabama,nloptr) as long as the canonical coefficients are treated as fixed-effects. This follows from the necessary identifiability constraints.
The reduced-rank approximated predictor slopes of a multivariate regression can now be plotted (with confidence intervals) using coefplot. Not available yet for quadratic effects.
Separate checks are put in place to warn users if the constraints on the canonical coefficients (orthogonality of the columns) have not converged.
Separate checks are put in place to warn users if the coefficients of a quadratic model have not converged
Canonical coefficients in ordination with predictors (num.RR,num.lv.c) can now be treated as random-effects using the ‘randomB’ argument. For the moment, all need to be either random or fixed, no mixing. Prediction intervals can be retrieved with the getPredictErr function.
An extended version of the spider dataset has been made available
Added an option to magnify the x-axis labels in coefplot
Site names present as row labels in the response data are now shown in the ordination plot
The order of the quadratic coefficients was wrong when num.RR, num.lv, and num.lv.c were all used in the same model.
Fixed a bug in the calculation of starting values for constrained ordination (num.RR) where the residuals were not re-calculated if num.lv.c>0
Fixed a bug in coefplot for when only one predictor was included in the model
Fixed a bug that would prevent using a gllvm with quadratic response model as starting values for another model
Changed import/export of various functions as requested in github issue #65
Various minor tweaks to the summary function
Structured row parameters are implemented, including a possibility for between or within group correlations for random row effects.
Constrained ordination model is implemented.
NB and binomial (with probit and logit) response model implemented using extended variational approximation method.
Quadratic latent variables allowed, that is term - u_i’D_j u_i can be included in the model using ‘quadratic = TRUE’. In addition, functions ‘optima()’, ‘tolerances()’ and ‘gradient.length()’ included.
Beta response distribution implemented using Laplace approximation and extended variational approximation method.
Tweedie response model implemented using extended variational approximation method.
Ordinal model works now for ‘num.lv=0’.
Residual covariance adjustment added for gaussian family.
Estimation of the variances of random slopes of the X covariates didn’t work properly when ‘row.eff = FALSE’ or ‘row.eff = “fixed”’.
Problems occurred in calculation of the starting values for ordinal model.
Problems occurred in predict() and residuals(), when random slopes for X covariates were included.
Problems occurred in predict() when new X covariates were given.
Problems occurred in predictLVs() for fourth corner models.
Structured row parameters are implemented, including a possibility for between or within group correlations for random row effects.
Constrained ordination model is implemented.
NB and binomial (with probit and logit) response model implemented using extended variational approximation method.
Quadratic latent variables allowed, that is term - u_i’D_j u_i can be included in the model using ‘quadratic = TRUE’. In addition, functions ‘optima()’, ‘tolerances()’ and ‘gradient.length()’ included.
Beta response distribution implemented using Laplace approximation and extended variational approximation method.
Tweedie response model implemented using extended variational approximation method.
Ordinal model works now for ‘num.lv=0’.
Residual covariance adjustment added for gaussian family.
Estimation of the variances of random slopes of the X covariates didn’t work properly when ‘row.eff = FALSE’ or ‘row.eff = “fixed”’.
Problems occurred in calculation of the starting values for ordinal model.
Problems occurred in predict() and residuals(), when random slopes for X covariates were included.
Problems occurred in predict() when new X covariates were given.
Problems occurred in predictLVs() for fourth corner models.