[Andrew Ng machine learning notes Chapter One Introduction: acquaintance Machine Learning

Search micro-channel public number: 'AI-ming3526' or 'this small computer vision' for more artificial intelligence, machine learning dry
CSDN: https://blog.csdn.net/qq_36645271
GitHub: https://github.com/ aimi-cn / AILearners

EDITORIAL

Andrew Ng teacher (Andrew Ng) machine learning course is the first course on Coursera, but also his classic. This course on-line from 2011 to 2017, this course has accumulated 1.8 million students worldwide, to many beginners AI provides a new way to learn.

This course can be said is the way to artificial intelligence, "the only way." I, as a graduate research NLP direction, even from the beginning I have not read the papers now mathematical meaning to the principles of machine learning and behind it there is a preliminary understanding, this course can not do without my help .

At the beginning of the note Special thanks Netease cloud classroom provides a learning platform for me, the course name to learn Andrew Ng machine .

At the same time thanks to subtitles and learning Dr Hong Hai wide offer of notes .

In my notes, I will refer to notes Dr. Wong and be understood in terms of their course, I wanted to be able to share together, learn together, and common progress.

This note only learn to share!

The first chapter: acquaintance Machine Learning

1.1 Welcome to the machine-learning courses

Machine Learning (Machine Learning, ML) is more than one field of cross-disciplinary , involving probability theory, statistics, approximation theory, convex analysis, algorithmic complexity theory and other subjects. How specializing in computer simulation or realization of human learning behavior to acquire new knowledge or skills, re-organize existing knowledge structures so as to continuously improve their performance . It is the core of artificial intelligence, is to make computers intelligent fundamental way, throughout all areas of application of artificial intelligence, it is mainly the use of induction, rather than a comprehensive interpretation. In this course, you will learn to machine learning this skill, and can achieve their own machine-learning algorithms.

In fact, every day we will deal with many machine learning algorithms

• Search in browser
• Taobao recommended for anything you might be interested in
• E-mail spam filtering system in

In this course, only learning algorithms and mathematical knowledge to understand the machine is not enough to make you able to solve practical problems, so we take the time to do some exercises to achieve these algorithms by having their own hands to understand its internal base management. Now that the data in the field of science python language is very hot, so I will use the notes python language to complete homework instead of course talking about Octave.

So why is it so popular machine learning now? that is because

• Machine learning is developed from AI Artificial Intelligence that is coming out of a field
• Machine learning is a computer developed a new feature

With traditional programming (similar to solving the shortest distance from A to B) are different, for example, a number of Web search, photo tags, filters spam and other work can not be the only human way to write code to implement simple logic, these functions is to let machine himself "learn" how to do it, this is not only machine in this course origin of the name, but also its charm!

Finally, we will introduce several other examples of machine learning:

• Data Mining

  One of the reasons machine learning is used for data mining is the growth of the network and automation technology, such as web click data, medical records, computational biology, various types of engineering.

• Programs written in humans can not

  Helicopter autopilot e.g., handwriting recognition, natural language processing (the NLP), computer vision.

• Personalized custom program

  Such as Amazon, Netflix, Taobao, iQIYI like recommendation system.

If the see show on the console 'Hello World!' Is the programmer's first lesson, then I think the phone open Taobao see accurate product engineers push algorithm is first class. I believe that when we complete completion of this course, the principle behind this precise push becomes no longer a mystery.

Machine learning is definitely the IT industry's most popular computer technology, let us take the pursuit of dreams and desire for AI-paid, together start our machine learning it!

---

1.2 What is Machine Learning

1.2.1 The definition of machine learning

Samuel arthur (1959) : When performing a particular program, so that the computer has the ability to learn the field of research.

Mitchell tom (1998) : a computer program from the experience of E (Experience) study, to solve a task T (Task) , to achieve the performance metrics P (Performance) , by P measured at T because of the experience on the performance of E increases. For example, in a checkers program, the experience E represents tens of thousands of chess under the program with their task T represents a play checkers, performance metrics P represents the probability of winning when playing checkers with the opponent.

1.2.2 a little exercise

Mail assume your e-mail program will observe you received is marked as spam. In such an Email client, you click the "Spam" button, some Email report as spam, the message does not affect the other. Based marked as junk mail, your e-mail program to better learn how to filter spam. Will, in this setting, E, P, T, respectively, which correspond to several?

To the message classification

Do you observe the mail marked as spam

The proportion of correctly classified messages

The correct answer are E, P, T. Through this example so that we can further understand the definition of machine learning - our performance on the task of T E after getting experience will improve performance P.

1.2.3 common machine learning algorithms

• Supervised learning: Human "teach" the computer to do a certain thing.
• Unsupervised Learning: humans make their own computer "learn."

1.3 supervised learning

1.3.1 Case 1: The rate forecast

By a straight line or quadratic fit to the existing data, and thus the predicted size of 750 square feet house prices. This is an example of supervised learning. ** As can be seen, supervised learning refers to the learning algorithm we give a set of data, this data set from the "correct answer" component. ** In the case of prices, we gave the house a range of data in the data set for each sample we give the right price, that is, their actual price, the purpose of supervised learning algorithm is to give more the correct answer, that is 750 square feet of the house should be priced.

Also, this is a regression of the forecast output a continuous value.

1.3.2 Case II: Breast cancer (malignant and benign)

In this example, the machine learning problem to be solved is to estimate the tumor is benign or malignant probability, the number 1 represents the tumor, the number 0 is not a tumor.

Also, this is a classification problem , namely the predicted output a discrete value.

In practice, however, problems to be addressed machine learning may not only be in one characteristic, in addition to the above example tumor size, age, and other characteristics may also, as shown in the following examples. All machine learning algorithms are able to handle only a two to three features, but the algorithm can handle an infinite number of features.

1.3.3 a little exercise

Suppose you run a company, you want to develop learning algorithms to deal with two issues:

1. Do you have a large number of the same goods, if you have thousands of pieces of identical goods to sell so you want to predict the next three months to sell many pieces?
2. Do you have many users, then you want to write a software to check each user's account. For each account, you want to determine whether they have been stolen?

That these two problems, they belong to the classification or regression?

问题一是一个回归问题，因为如果我有数千件货物，我会把它看成一个实数，即一个连续的值。因此卖出的物品数，也是一个连续的值。

问题二是一个分类问题，因为我可能会用 0 来表示账户未被盗，用 1 表示账户曾经被盗过。所以我们根据账号是否被盗过，把它们定为0 或 1，然后用算法推测一个账号是 0 还是 1，因为只有少数的离散值，所以我把它归为分类问题。

1.3.4 监督学习总结

监督学习，核心在“监督”二字，基本思想是，我们数据集中的每个样本都有相应的“正确答案”，即都被标记。再根据这些“正确答案”对新样本进行“监督”，最后得到预测结果。就像房子和肿瘤的例子那样。同时还介绍了回归问题，即通过回归来推出一个连续的输出，这里还介绍了分类问题，其目标是推出一组离散的结果。

1.4 无监督学习

1.4.1 无监督学习与监督学习的区别及其概念

监督学习的数据集，如上图左1表所示，其中每条数据都已经被标注，例如一个肿瘤是良性或恶性。所以，对于监督学习里的每条数据，我们已经清楚地知道，训练集对应的正确答案。

无监督学习，核心在“无监督”这三个字，我们数据集中的每个样本都没有相应的“正确答案”，即都未被标记，我们将这些数据交给算法，并让算法为我们从中找出某种结构。无监督学习的数据集与我们之前看到的不一样，如上图右1所示，所有数据都没有被标注，即无监督学习中没有任何的标签或者是都具有相同的标签。所以对于一个没有任何标签的数据集，无监督学习算法可以判定该数据集包含两个不同的簇（Cluster）。

同时，这是一个聚类问题。

1.4.2 无监督学习算法的应用

• 新闻网站。新闻网站会用聚类算法将每天收集到的成千上万条的没有任何标记的新闻组合成一个个新闻专题。
• 基因组学。通过聚类算法，在没有提前告诉这个算法什么样的个体是什么种类的情况下，把不同的个体归为不同的类。
• 组织大型计算机集群。
• 社交网络分析。
• Market segments.
• Astronomical data analysis.
• Cocktail party problem (Interested students can go to class 4 unsupervised learning to learn more).

1.4.3 a little exercise

In the following example, which is suitable for use unsupervised learning calculation method to solve?

1. E-mail data is given / not spam label, training an e-mail filtering system.
2. Given a series of new articles found on the Internet, they will be assigned to the same content of the article group.
3. Given a set of customer data, automatically segments, and different customers into different groups.
4. Given a set of data sets for diabetes patients, and then predict whether a new patient suffering from diabetes.

Obviously, the previous example, we can easily come in conjunction with 2,3 suitable for unsupervised learning algorithms to solve, and 1,4 suitable for supervised learning algorithm to solve.

---

AIMI-CN AI learning exchange group [1015286623] For more information on AI, scan code plus group:

sharing technology, fun in life: Welcome to our public number, push AI Series News articles per week, welcome your interest!