Algorithms for Optimization Hardcover – Illustrated, 12 March 2019

Este libro simplemente es increíble. Presenta temas muy avanzados de una forma clara y practica. Incluye los códigos de julia.

This book was mind blowing.It covers around 100 different algorithms for optimization. Probably more; I didn't count thoroughly.It describes algorithms and concepts with incredible clarity and extreme concision.It builds progressively from simple to complex.It provides all the background information needed beyond a basic calculus class and some basic background dealing with matrices and vectors.It provides code snippets written in Julia of all the algorithms.It includes exercises and answers. Other examples are presented throughout the text.It provides resources online that run using Jupyter notebook with a Julia kernel.This book refreshed my memory and introduced me to so many topics. In particular, I found the sections on automatic differentiation, computational graphs, optimization under constraints, multiobjective optimization, surrogate models, sampling plans, and expression optimization to be enlightening and in some cases revolutionary to me. Like, OMG, you can do that? Over and over I thought, "I'll just skip this section. It seems irrelevant to what I need to learn." And each time I thought that, I'd start reading the section and would get hooked. Almost every section was highly relevant and provided building blocks for a deeper understanding. The book clarified so may ideas for me: function approximation, Lagrange multipliers and their extensions, duality, Pareto optimality, uses of quasi-random sequences, surrogate models, and probabilistic grammars. All of these ideas will be useful in my current projects.Julia was new to me. This language seems to be able to represent many loop structures and iteration processes in extremely compact form. Downloading and installing it and all other Julia modules used by the book was straightforward (except the Vec package needed a bit more sleuthing to get).Don't be fooled, though. This is an introductory text, and based on the preface, it appears to be intended for undergraduate-level courses. You will not find proofs of the results presented in the book - that is not the goal of the book. Margin notes provide relevant references from the primary (and secondary!) literature. For example, I had to look up more about probabilistic prototype trees and learning algorithms for these structures; it was a snap to find the relevant primary literature. The book's real strength is in the sheer number of algorithms described.Despite the comprehensive coverage, not all topics I was expecting were covered. I was hoping for something about expectation maximization and other latent variable methods. I also was hoping for more information about optimization with decision trees. Also, MCMC was missing although some Monte Carlo approaches were described; usually, the book advocated other methods over Monte Carlo approaches for more efficient optimization. Granted, this book is not intended as a machine learning book that might cover these missing topics in more detail. (BTW, the methods in the book can certainly be applied to machine learning problems. )The book sort of just ends. A final synthesis chapter that provides tables of the strengths, weaknesses, and areas of applicability of all the methods covered in the book, or a chapter outlining current challenges and areas of research, would be icing on the cake. The reader must make this synthesis themselves. Strengths and weaknesses are covered during the exposition of the various approaches, so this synthesis could be done with some discipline on the part of the reader.

It's a great book which guides carefully through the different level of optimization. Examples and exercises are useful and can be easily adapted for own software projects.

This book is worth reading. As someone already pointed out, it is a huge gallery of algorithms of all kinds. Explanations and derivations are sometimes too brief, however it covers a variety of modifications, some of them being very niche. Thus if you consider the book as a survey of different strategies or you find yourself wandering on you problem unable to decide or even find good systematic way, chances that you find an inspiration here are indeed very high. I use this book as a first look, but due to lack of depth, I always navigate elsewhere to gain more. This way I discovered whole new realm of Bayesian approach and was blown away. Of course I ended up with Garnett book on that, but you get the point. This book gives you ideas and I highly recomend purchasing it.

Julia is not as common as other programming languages like Python, Matlab, making examples harder to follow for me.