Teacher Wu Enda machine learning course notes 01 introduction

1 introduction

1.1 welcome

Machine learning algorithms are applied to every aspect of life , Such as web search 、 Photo tag 、 anti-spam .

The origin of machine learning 、 Popularity and development

Machine learning originated from AI.

To create intelligent machines , Traditional programming can only let machines do some basic things , For example, find the shortest path between two points . But if you let the machine learn by itself , You can do more interesting things , As the above three examples .

therefore , Machine learning is a new function of computer development , Now it has involved many fields of industry and basic science .

Examples of machine learning applications

• data mining
  • Turn electronic medical records into medical knowledge
• Tasks that cannot be programmed manually
  • Handwriting recognition
• Personal customization program
  • Custom recommendations
• Understand the human learning process and understand the brain

summary

Machine learning originated from AI, Generally speaking, let the machine learn by itself , The purpose is to make the machine more intelligent .

At present, machine learning has been widely used .

1.2 What is machine learning

Definition of machine learning

A more formal definition of machine learning is ： A program from experience E Middle school learning , Solve the task T. use P Measure performance , adopt P Determination , The program is solving the task T Performance is due to experience E And improve .

For example, for alpha Go, Experience E It's the experience of the program practicing itself many times , Mission T Is playing chess , Performance metrics P Is its probability of winning when playing chess with a new opponent .

The types of machine learning algorithms

• Supervised learning （ Commonly used ）
  • Teach the computer to complete the task
• Unsupervised learning （ Commonly used ）
  • Let the computer learn by itself
• Reinforcement learning
• Recommendation system

Learning how to apply algorithms is very important

This course will spend a lot of time giving practical suggestions and practical experience of applying the learning algorithm learned , Make readers learn how to use tools .

summary

One definition of machine learning is ： A program from experience E Middle school learning , Solve the task T, use P Measure performance . adopt P Determination , The program is solving the task T Performance is due to experience E And improve .

Machine learning algorithms are mainly divided into supervised learning and unsupervised learning .

In addition to learning the algorithm itself , Learning how to practice algorithms is equally important .

1.3 Supervised learning

The definition of supervised learning

Supervised learning refers to giving the learning algorithm a data set , It contains the correct answer , The task of the algorithm is to give more correct answers .

For example, for the prediction of house prices , Given a house price data set , The area and actual selling price of each sample in the known data set , Then use the learning algorithm to get the selling price of the new house .

For example, for predicting whether breast cancer is benign , Given a breast cancer dataset , The tumor size and benign or not of each sample in the data set are known , Then use the learning algorithm to get whether the new tumor is benign .

The return question

In more technical terms , The above problem of predicting house prices is called regression problem .

Regression problem refers to the result of predicting a continuous value , The price of the house .（ House prices can usually be regarded as real numbers , Continuous value .）

Classification problem

In more technical terms , The above problem of predicting whether breast cancer is benign or not is called the classification problem .

Classification problem refers to the result of predicting a discrete value , Benign or malignant .

Of course, there are sometimes more than two possible results in classification problems .

In this case , Only one feature is used （ attribute ） To make predictions , There may also be multiple features in other machine learning problems （ attribute ）.

summary

Supervised learning refers to giving the learning algorithm a data set , It contains the correct answer , The task of the algorithm is to give more correct answers .

Regression problem refers to the goal of predicting the output of a continuous value .

Classification problem refers to the problem that the goal is to predict the output of a discrete value .

1.4 Unsupervised learning

The definition of unsupervised learning

Different from supervised learning data , Data in unsupervised learning has no labels . I don't know what the data is , I don't know what types of data , I don't know what I can do with data sets , Instead, we need to find some structure .

clustering algorithm

Clustering algorithm is an unsupervised learning algorithm , Is to divide the data into several different clusters .

Application, for example,

• News sub group
• Depending on the genes , Cluster individuals into different groups
• Group friends
• Market classification （ Divide customers into different market segments ）

Cocktail party algorithm

Cocktail party algorithm is also an unsupervised learning algorithm , It refers to the given mixed sound signal , Separate the independent sound signals of everyone who speaks at the same time in the cocktail party .

summary

Data in unsupervised learning has no labels , The algorithm needs to find some structure from it .

Clustering algorithm and cocktail party algorithm are unsupervised learning algorithms .