# How to choose which machine learning subjects to skip for later?

First, let me preface by admitting that I don't know if this question is off-topic for this site. It is important enough to me that I am willing to risk the inevitable backlash in case it is off-topic.

Question: Given that they have no context for which topics are important or not, how can a machine learning novice choose which parts of a very long introductory textbook to focus on, and which parts to leave for later when their knowledge is more advanced?

I have no formal training in machine learning, so I wanted to acquaint myself with some of the concepts, e.g. what is CART or boosting. However, there does not seem to be any introductory textbook (at any level of mathematical sophistication) which is short. For example:

• Bishop, Pattern Recognition and Machine Learning (700+ pages)
• Goodfellow, Benigo, Courville, Deep Learning (800+ pages)
• Hastie, Tibshirani, Friedman, Elements of Statistical Learning (700+ pages)
• Murphy, Machine Learning (1000+ pages)
• MacKay, Information Theory, Inference, and Learning Algorithms (600+ pages)
• Raul Rojas, Neural Networks (500+ pages)

Even if I choose only one of these books as the one I want to read to the exclusion of all others, it seems unlikely that reading that one book cover to cover would be a very productive strategy.

Here are some ideas I had which don't seem to work:

• Skip the sections which are obviously inessential advanced material: I don't know how to differentiate those sections from the sections containing essential basic material.
• Start with the sections I'm most interested in: I don't know which topics I'm interested or not interested in, because I don't even know what the topics are about (again, no background).
• Skip the parts covering mathematical background I'm already familiar with: True, most of it will be redundant. But, based on skimming, it seems that the authors use it and think about it in a very different way than that which I'm accustomed to, and there are one or two new things (e.g. multinoulli or Dirichlet distributions) which never came up in my education.

My goal is to ultimately be able to select about ~400 pages of one, maybe two, books to focus on understanding.

• There is An Elementary Introduction to Statistical Learning Theory, about 200 pages. – hellpanderrr Jun 3 '18 at 23:01
• And An Introduction to Statistical Learning with Applications in R, ~400 pages – hellpanderrr Jun 3 '18 at 23:05
• @hellpanderrr The first one is new to me, and looks like a very good recommendation; I appreciate it! A professor also forwarded me to here: ciml.info (< 230 pages) and here arxiv.org/abs/1803.08823 (< 120 pages). – Chill2Macht Jun 11 '18 at 1:20