Machine learning Getting Started Guide

The predecessor of the network said: machine learning is not an isolated algorithm piled up, want to look like "Introduction to the algorithm" to see machine learning is an undesirable method. There are several things in machine learning that have been running through the book, such as the distribution of data, the maximum likelihood (and several methods of finding extreme values, but this is more mathematically), the tradeoffs of deviations, variances, and the knowledge of feature selection, model selection, mixed models, etc., such as bricks, Cement is the same as the machine learning inside the algorithm. To really learn these algorithms, it is important to calm down the basic knowledge to make clear, to truly understand and achieve a variety of machine learning algorithms.

There are a few things to learn about:

1. Online Courses

Tai Lin Xuan Tian teacher on the Coursera on the machine learning Cornerstone is a good introduction to learning theory, he in the machine learning techniques, but also in the introduction of a number of common algorithms and supplemented by theoretical explanations, originally the two classes in Taiwan is one of the big. If Learning machine learning, to get started, personally think is preferred.
Alternative online courses have Wunda courses that are more accessible than the former.

2. Textbooks

Book aspects do not suggest directly chew Zhou Zhihua professor's machine learning (commonly known as Watermelon book, because most of the space to the watermelon classification and other issues as the background, hence the name), there is a lot of math, the new words feel relatively easy to get lost in the details. "Machine learning Combat" is used to get started, personally feel good, although many people spit groove among the code quality. Watermelon Book and professor Hangyuan Li's "Statistical learning method" can be used in the desk, often read new. The classical textbooks abroad have not been read, not evaluated. In addition, the collective intelligence programming >< Data Analysis actual combat >< Artificial intelligence: A modern approach to broaden the field of vision is a rare good book, do not miss easily.

3. Tool Software

Learning while validating, more efficient. Some mature machine learning framework, can make us more convenient in learning to obtain perceptual cognition. In this regard, the first Scikit-learn python system, ease of use not to say, the document is also moving to No. Alternative to spark in the Mllib and Mahout, the pros and cons do not repeat, a check will be clear. If it is a statistically cheese, using R or MATLAB is of course also very good.

4. Mathematical

Statistics/probability/optimization (convex optimization), in which all the knowledge deep like the sea, and really short. But the human energy is limited, to the beginner can temporarily slow down, focus on fire a bit, first of all the machine learning Knowledge system has a general understanding, in the mathematical knowledge continues to be strong. Machine learning is rooted in mathematics, and there is no relevant reserve to go far.

5. Actual combat

Kaggle and Ali Tianchi can try to play, find some simple toy topic, do not ask for results, can be familiar with the entire analysis process, the application of learning on good, want to become Daniel, slowly, relatively fast.

If you're already working, it's good to try to find the app at work, and it will make you work better!

　　　　

Machine learning Getting Started Guide