High-Performance Open-Source Archive
Version increased to 1.0.0 to reflect publication of Youngman (2022, JSS, ).
References to Youngman (2022, JSS, ) added, where appropriate.
data is now
properly detected.plot() for an evgam object now calls
mgcv::plot.gam() to plot smooths (with thanks to Debbie
Dupuis for triggering this). plot() no longer has the
addMap option, for adding map outlines via
maps::map(); instead using one-figure devices with
maps::map() separately is recommended.
Calculations of log(|S|_+) for penalty matrix S now fully implements Wood (JRSSB, 2011(73)1, Appendix B).
Calculations of log(|H|) for Hessian H now use diagonality simpifications; see Wood (book: GAMs in R 2nd ed. (2017) pp. 286).
The Fremantle data from package ismev have been added, and are
used for examples. Usage is data(fremantle), as in
ismev.
colplot() adds the option to add a legend, which
defaults to FALSE.
logLik.evgam() now returns an object of class
'logLik', allowing, e.g., AIC() and
BIC() to be used.
extremal0() has gone, as extremal() can
now do the same.
evgam()’s trace argument now allows -1, which
suppresses any information on the console.
Negative response data now work okay with
family = "ald".
evgams()’s formula argument may have smooths and
parametric-only terms in any order. (Previously, smooths had to come
first, so formula = list(response ~ s(), ~ 1, ~ s())
broke.)
predict.evgam(object) with
missing(newdata) only gave one set predictions for
object$data. It now gives predictions for all rows of
object$data (as it should).
plot.evgam() now has informative y-axis labels for
one-dimensional smooths.Compilation flag with clang++ in gradHess.cpp addressed.
simulate.evgam() correctly labels variables for
family = "response".
Need mirroring services?
Contact our team at info@vpspulse.com.
Mirror powered by VPSpulse
Infrastructure sponsored by VPSPulse & Secure Payments by ArionPay.