Machine Learning Pocket Reference: Working with Structured Data in Python

Machine Learning is a large domain and a book covering this topic needs to choose carefully what to cover. In Machine Learning Pocket Reference, the author chooses to focus on processing structured data. This means he avoids discussing neural network libraries such as TensorFlow or Natural Language Processing tools like spaCy or NLTK. This conscious decision means he can focus on clear and detailed code examples for solving traditional classification or regression problems using scikit-learn (and other python tools). Each chapter uses concise code samples to walk through how to use many different python packages to work through the multiple steps of a typical machine learning problem. This book is best for someone that has a little bit of exposure to python, pandas and scikit-learn and wishes to learn how to use these tools effectively. It also provides a very good introduction to about 36 other python libraries commonly used in the data science field.If you find yourself looking for quick reminders of how to use functions or are interested in multiple approaches to solving a python data science problem, this book will be a great addition to your bookshelf (real or virtual).One quick note about the size of the book. It is a quick reference so it is suitable for carrying around in your laptop bag. I was a little surprised about the size, so I am attaching an image so you can get a sense for the dimensions of the book.

As Matt (the author) states in the intro to this book, it will not teach you "from scratch" data science or machine learning. While not suitable for one's first exposure to the material, it serves an excellent companion piece once you have some foundational data science/ML knowledge and a little bit of Python knowledge (pandas will help too).The book provides a thorough, example-driven treatment of every major step in a DS/ML project, from data cleaning to model evaluation. There's also coverage in terms of many common model scenarios (classification and regression) and the main models within each of those buckets (e.g., Logistic Regression for Classification and Linear/Random Forest for regression). While the book covers a lot of territory relatively speaking, the focus is on practical implications/takeaways versus going through a lot of academic background/boilerplate material. This approach gives the book a nice balance of levity and economy in its delivery.An aspect of the book that I appreciated is that it is very well segmented into terms of individual topics. Often points of interest have their own short section and are not buried in a forest of dense paragraphs. For example, if I wanted to look up something on how to deal with missing data or pre-processing steps, these have separate sections with example code provided.A pleasant surprise from this book was that it exposed me to a somewhat recent data science visualization library Yellowbrick. It's a very handy and easy-to-use library for visualization a lot of common data science charts (e.g., confusion matrix, residual versus predicted value). There are other examples of new data science visualization tools that the reader will be exposed to that I have not seen covered elsewhere (e.g., Shapley).Overall, this is a valuable companion piece either to go through end-to-end or to use as a reference when going through your own data science projects.

good worked out examples

Very good overview of a theme with examples.

Aun tratándose simplemente de un manual de bolsillo, introduce el uso de bibliotecas recientes como yellowbrick que no aparecen en libros mucho más profundos sobre la materia. Es un librito interesante, que está bien tener para refrescar conceptos de vez en cuando con ejemplos claros, sencillos y que funcionan perfectamente. Si eres completamente nuevo en la materia, no te servirá de gran ayuda. Es más bien para lectores que ya tienen algún conocimiento sobre el mundo del aprendizaje automático. Como complemento de otras lecturas está muy bien.

Great reference with very useful cases and applications. It lacks deep learning but it is otherwise very comprehensive. I wish there will be a similar one dedicated to deep learning.