Diving into the second edition of Simon Wood's GAM book felt like upgrading from a bicycle to a sports car. The restructured chapters flow logically, and the new adaptive smoothing section saved me hours of frustration when working with unevenly sampled environmental data.

What really shines is how Wood balances theory and practice. The mgcv package examples aren't just code dumps - they're thoughtfully integrated with explanations that helped me debug my own ecological models. I particularly appreciated the expanded coverage of exponential families when modeling disease progression data.

The book isn't perfect - I do miss that handy smoothing bases comparison table from the first edition. And yes, the QR decomposition sections made my eyes glaze over until I pushed through to the meatier chapters. But when I finally implemented a location-scale model for my heteroskedastic sales data, those dense theoretical sections suddenly clicked into place.

This isn't a gentle introduction - you'll need solid GLM and linear algebra foundations. But for practitioners who need to move beyond textbook linear models, it's become my most dog-eared reference. The functional data analysis additions alone justified my purchase when working with sensor time series last quarter.