Moroney's AI and Machine Learning for Coders review spans 20 chapters

The review on Qiita is notable for how it frames the book as a single line from model creation to delivery. The 20 chapters split into two parts: Part I (chapters 1–11) builds foundations, image, natural language, and time-series models, while Part II (chapters 12–20) covers TFLite, Android, iOS, TensorFlow.js, Serving, and ethics. The reviewer notes that the basic pattern—data preparation, model definition, compile, train, evaluate—stays constant, with only the input and output shapes changing.

The review highlights a consistent emphasis on questioning accuracy numbers. For example, it points out that adding dropout lowers training accuracy while narrowing the gap with validation, and that normalized losses can look smaller than they are unless evaluated on actual values.

The book also treats deployment constraints as design inputs: HTTP connection limits and memory in browsers, model size and type conversion on mobile, and version management on servers each shape how models are built. The review notes that the book's code assumes TensorFlow 2.x and may change with later versions, and that Part II only introduces each platform rather than going deep.