You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
In all tutorials and relevant lecture slides, provide links to mvgam vignettes
Day 1
Link to the ESA talk rather than the Stat Rethinking Categories and Curves
Skip over ETS and ARMA slides in lecture
Skip through the Beta example as well
Introduce the portal example when first talking about random effects; then pause, switch to live coding and show how to do this in mgcv (provide syntax to do it in brms, mvgam and glm as well); show how to use marginaleffects to make hindcast predictions; emphasize how to create smooth plots; emphasize how to take realisations of fitted functions so they can be used for more custom plots or analyses
Switch to live coding for the yearly smooth, again showing in mgcv; use gratia to show the basis; extract residuals and plot against another predictor to start building in the idea of a workflow
Link to Nishan's paper in the tutorial and include the cheatsheet pdf in the supplied files
Day 2
Link to the Oceania EcoForecasting seminar
Skip over the tscount AR example slides
For the dynamic Beta GAM, pause and switch to live coding; show how to do something similar in mgcv with the RW MRF basis; talk about tensor products and introduce the idea of ti() decompositions to help build strategies (link to some of Gavin's Stackoverflow descriptions and the gam.models help page as well)
Skip over the enforcing stationarity Student T example
In the dynamic coefficient example, switch to live coding; show how to do something similar in mgcv with the RW MRF basis; mention how one could use the same principles for spatially varying coefficients
Use Hilbert GP in dynamic coefficient mvgam example
In tutorial fix typo: "except the length scale was changed to 3" should be "except the length scale was changed to 4"
Subtitles should use "GAM" rather than "GLM"; i.e. "A standard Poisson GLM" should be "A standard Poisson GAM"
Link to MRFtools
Include the cheatsheet pdf in the supplied files
Day 3
Link to the forecasting vignette and to Juniper's paper on forecast evaluation
Skip over expectations entirely and reduce down the conditional predictions part
At the types of predictions, switch to live coding and show something cool / interesting (distributed lags in mgcv)
Same at probabilistic forecast evaluation; perhaps show a strategy to estimate a decaying effect of some treatment (time since treatment example, or time-varying dispersion with a distributional model)
Remove stochastic trend extrapolation completely, and talk about loo_compare instead; relate back to the cheatsheet
Include the cheatsheet pdf in the supplied files
Day 4
At hierarchical dist lags, switch to live coding and show something cool / interesting (phylogenetically informed intercepts and nonlinear functions in mgcv; illustrate prediction by excluding certain terms to show how the trend is built of additive functions)
At multivariate forecast evaluation, switch to live coding and show time-varying seasonality in mvgam
Day 5
Pick a more simple and useful dataset for groups to analyse
Be sure to create PDFs of all lectures again, once finalised
The text was updated successfully, but these errors were encountered:
In all tutorials and relevant lecture slides, provide links to
mvgam
vignettesDay 1
mgcv
(provide syntax to do it inbrms
,mvgam
andglm
as well); show how to usemarginaleffects
to make hindcast predictions; emphasize how to create smooth plots; emphasize how to take realisations of fitted functions so they can be used for more custom plots or analysesmgcv
: https://stats.stackexchange.com/questions/618715/building-the-right-gam-model-struggling-with-the-jump-from-lmer/618760#618760mgcv
; usegratia
to show the basis; extract residuals and plot against another predictor to start building in the idea of a workflowpdf
in the supplied filesDay 2
tscount
AR example slidesmgcv
with the RW MRF basis; talk about tensor products and introduce the idea ofti()
decompositions to help build strategies (link to some of Gavin's Stackoverflow descriptions and thegam.models
help page as well)mgcv
with the RW MRF basis; mention how one could use the same principles for spatially varying coefficientsmvgam
exampleMRFtools
pdf
in the supplied filesDay 3
mgcv
)loo_compare
instead; relate back to the cheatsheetpdf
in the supplied filesDay 4
mgcv
; illustrate prediction by excluding certain terms to show how the trend is built of additive functions)mvgam
Day 5
Be sure to create PDFs of all lectures again, once finalised
The text was updated successfully, but these errors were encountered: