目录

●Chapter 1 Introduction
1.1 Generalized additive models (GAMs) and mgcv
1.2 Black Smoke (BS) in the UK
1.3 Daily log Black Smoke (logBS) Dataset
1.4 The Network for Black Smoke monitoring
1.5 Previous modelling of log Black Smoke
1.6 Regression computing for large datasets
1.7 Integrated nested Laplace approximation (INLA)
1.8 About this book
1.9 Notations and abbreviations
Chapter 2 Penalized regression splines and additive models
2.1 Penalized cubic regression spline
2.2 Penalized thin-plate regression spline
2.3 Penalized tensor product regression spline
2.4 Canonical form of penalized regression splines
2.5 Bayesian interpretation of smoothing
2.6 Smoothness selection via REML
2.7 Additive models
Chapter 3 Additive models in practice: Simulated examples
3.1 Classic example with independent errors
3.2 Time series example with AR(1) errors
3.3 Visualization of a tensor product spline
3.4 Model checking
3.5 Summary
Chapter 4 Preliminary modelling of log Black Smoke (logBS)
4.1 Daily logBS at Manchester 11
4.2 Motivating spatial-temporal models
4.3 Spatial-temporal model of yearly mean logBS
4.4 Spatial-temporal model of daily logBS in 1967
4.5 Summary
Chapter 5 GAM computation for large datasets: A review
5.1 Preliminaries on matrix computations
5.2 Model matrix reduction for additive models
5.3 REML estimation for additive models
5.4 Estimating GAMs via performance iteration
5.5 Summary
Chapter 6 Optimizing GAM computations for HPC
6.1 Improving existing computational engine
6.2 Using optimized BLAS for HPC
……
9.4 Experimenting BAMBOOS on daily logBS model
9.5 Summary
Chapter 10 Spatial-temporal modelling of the UK's daily logBS
10.1 Daily logBS model
10.2 Modelling results
10.3 Spatial-temporal prediction
10.4 Bias of spatial-temporal prediction
Chapter 11 Discussion
11.1 Big sparse GAMs
11.2 Suppressing spatial-temporal extrapolation
Appendix A Hardware information
Appendix B Simulations on spatial-temporal prediction bias
Bibliography