Skip to content

Latest commit

 

History

History
130 lines (78 loc) · 4.36 KB

chapter4_12.md

File metadata and controls

130 lines (78 loc) · 4.36 KB
type
slides

Doing more with GAMs

Noam Ross Senior Research Scientist, EcoHealth Alliance

Notes: Congratulations! You've have successfully completed this course on Generalized Additive Models. Let's quickly review what you've learned, and point you towards a few resources for you to use as you apply these models in your future work.

Course review

Chapter 1

  • GAM theory
  • Fitting GAMs
  • Mixing linear and nonlinear terms

Chapter 2

  • Interpreting GAMs
  • Visualizing GAMs
  • Model-checking and concurvity

Notes: In the first chapter of the course, you learned the basic theory of how smooths are constructed and how to assemble GAMs from multiple smooths and linear terms. In chapter two, you learned how to interpret GAM model outputs, plot partial effects to check your models for problems and how to fix them.

Course review

Chapter 3

  • 2-D Interactions and spatial data
  • Interactions with different scales
  • Continuous-categorical interaction

Chapter 4

  • Logistic GAMs
  • Plotting logistic outputs
  • Making predictions

Notes: Then, in chapter three, you expanded to building and visualizing GAMs with interactions of multiple variables, including geospatial GAMs and GAMs with continuous-categorical interactions. Finally, in chapter 4, you learned how to use logistic GAMs for binary classification and prediction. Altogether, you've created a toolbox for using GAMs to model many different types of data and problems.

Now I'll point you to a few more resources for using GAMs in your work.

GAMs and the Tidyverse

library(broom)

augment(gam_model)
tidy(gam_model)
glance(gam_model)
library(caret)

train(x, y, method = "gam", ...)

Notes: In this course we did not make use of tools from the popular set of R packages known as the tidyverse, but GAMs work readily with tidy tools.

If you are familiar with the broom package, you'll know of general functions like augment(), tidy(), and glance(). All these functions work well with GAMs, giving you tidy ways of inspecting, evaluating, and predicting from your models.

Similarly, if you use the caret package for predictive modeling, you can pass "gam" to the method argument and caret will fit models with the mgcv package.

Other types of smooths

?smooth.terms

Notes: As you continue to use GAMs, you'll discover mgcv has many additional capabilities. The package also has many extended help files on specific topics that will be useful to explore.

First, there are many additional types of smooths beyond the ones we have used here. These can be useful in specific situations, such as geospatial modeling or for seasonal time series. You'll find these described in the ?smooth.terms help file.

Other types of outcomes/distributions

?family.mgcv

Notes: If you have taken a course in generalized linear models, you probably know that there are many types of outcomes, such as count data, which they can model. These can be fit with GAMs, as well. mgcv also has an extensive collection of outcome distributions above and beyond those available in most GLM packages. You'll find them described in the ?family.mgcv help file.

Variable selection

?gam.selection

Notes: mgcv has tools for variable selection when model building. You can learn about these in the ?gam.selection help file.

Complex model structures

?gam.models

Notes: Finally, there are other options in mgcv for alternative or more complex model structures, such as mixed effects. The ?gam.models help file has an overview of these topics and will point you to more documentation. There are also a number of links to resources on this topic in the Reference section of this course

Thank You and Keep Learning!

Notes: Thanks for taking this course on nonlinear modeling in R with GAMs. I hope you find these flexible and powerful tools useful in your work.

In the final section of this course you will find reference code for re-use and links to more resources.