[machine learning] - Introduction to vernacular and explanation of terms

---

Preface

Machine learning and artificial intelligence , I always think it's mysterious and tall , I often hear that , But it has not been put into practice because of various mathematical concepts .

however , If you don't do full-time relevant post development , Is it feasible to run some learning programs by yourself ？ There are a lot of frameworks now , Even if it's not enough , It's also good to know what you can do .

Starting with Zhou Zhihua's 《 Artificial intelligence 》 A Book , Just read the beginning , I think it's very good . Start with examples and gradually deepen , Made the following notes .

One 、 Step by step with common examples to explain what machine learning is

Start with the example of picking watermelon ： Why is the color green 、 Root curl 、 The sound of knocking is cloudy , You can judge that it is a good ripe melon ？

Because we ate 、 I've seen a lot of watermelons , So based on the color 、 roots 、 We can make a fairly good judgment on the characteristics of knocking .

** transition , Lead to learning experience ：** Allied , We know from our previous learning experience , Worked hard 、 Figured out the concept 、 Finished the homework , Naturally, you will get good results . It can be seen that , We can make effective predictions ？ Because we have accumulated a lot of experience , And through the use of experience ? Can make effective decisions about new situations .

** Further, machine learning ：** If computer science is about " Algorithm " Knowledge of , So similar , It can be said that machine learning is about " Learning algorithms " Knowledge of .

Two 、 Learn some related terms through the analogy of watermelon

1. Learn in the form of data tables

Summarize the examples of watermelon into the following table ：

Serial number	Colour and lustre	roots	Knock sound
1	dark green	Curl up	Murmur
2	It's dark	Curl up a little	Dull
…

Comparison table , Know some related terms ：

• ** Data sets ：** The record set of the whole table .
• Example (instance) Or sample （sample）： Each record is about an event or object （ Here is a watermelon ） Description of , Called an example (instance) Or sample （sample）. Sometimes the whole dataset is also called a " sample " Because it can be seen as a sample of the sample space , It can be judged by the context " sample " Single sample or dataset .
• ** attribute (attribute) Or characteristics (feature)：** In the table “ Colour and lustre ”、“ roots ”、“ Knock sound ”.
• ** Property value (attribute va1ue)：** In the table “ Colour and lustre ”、“ roots ”、“ Knock sound ” Corresponding value .

2. Remember the coordinate system

For a single record , With “ Colour and lustre ”、“ roots ”、“ Knock sound ” The three attribute identifiers are shown in the figure below ：

Each attribute acts as a coordinate axis , It forms a three-dimensional coordinate system . Remember the coordinate system , I hope I haven't returned it all to the teacher .

The space formed by attributes is called **“ Property space ” (attribute space) “ sample space ” (sample space) or " input space "**, That is, the box in the figure .

Of course , In fact, a sample （ watermelon ） There must be more than these three attributes , This is just an example . Each attribute represents a coordinate axis , That will form a d Dimensional space ,d Is the number of attributes of the sample .

Each watermelon can find its own coordinate position in this space . Because each point in space corresponds to a coordinate vector , So we also put … An example is called a **“ Eigenvector ” (feature vector).**

3. Some terms related to training

The process of learning models from data is called **“ Study ” (learning) or " Training " (training)**;

This is done by executing a learning algorithm . The data used in the training process is called **“ Training data ” (training data)** ;

Each of these samples is called a The training sample " (training sample),;

The set of training samples is called " Training set " (training set).

The learned model corresponds to some potential law about data , So it's also called " hypothesis " (hypothesis);

The underlying law itself , It is called a **“ The truth " or " real ” (ground-truth)** ;

The learning process is to find out or approach the truth . This book sometimes calls models " Learner " (learner) , It can be regarded as the instantiation of learning algorithm in given data and parameter space .

Training requires more than the attribute information of the sample , Samples are also required " result " Information , for example " (( Colour and lustre : dark green ; Rooty two curled up ; Knock sound = Murmur ), Good melon )" . Here's information about the example results , for example " Good melon ", be called Mark (labe1);

With an example of tag information , It is called a **“ Examples ” (example)**. Here's the picture

If we want to predict discrete values , for example " Good melon " “ Bad melon ”, Such learning tasks are called **“ classification ” (classification)**;

If you want to predict a continuous value ? For example, watermelon maturity 0.95 0.37, Such learning tasks are called **“ Return to ” (regression)**.

Yes, there are only two categories **“ Two classification ” (binary classification)** Mission , One of the classes is usually called Just like （positive class）, The other class is **“ Anti class ” (negative class);**

When multiple categories are involved , It is called a **“ Many classification ” (multi-class classificatio）** Mission .

The ability of learning model to apply to new samples , be called " generalization " (generalization) Ability . The model with strong generalization ability can be well applied to the whole sample space .

3、 ... and 、 Hypothetical space

In the mirror of mathematical axioms , Based on a set of axioms and inference rules, the corresponding theorems are deduced , This is the interpretation ; and " Learn from examples " It's obviously a process of induction , So it's also called " Inductive learning " (inductive learning) ..

We can think of the learning process as a process in which all assumptions (hypothesis) The process of searching in the composed space . Here's the picture

How many possibilities are there , It's a permutation . In reality, we often face a lot of hypothetical space , But the learning process is based on the limited sample training set , therefore , There may be multiple assumptions that are consistent with the training set , That is, there is a consistent " Suppose the set ", We call it **“ Version space ” (version space)**.

Four 、 Generalize preferences

For Graphs 1.2 Watermelon version space , Corresponding ( Color mouth = dark green ; roots = Curl up ; Knock sound = Dull ) This new melon , If we were to use the " Good melon <->( Colour and lustre =* )^{( roots = Curl up )}( Knock sound =*), Then the new melon will be judged as a good melon , And if two other assumptions are used , Then the result of judgment will not be good . If only the watch 1. Training samples in , It is impossible to determine which of the above three assumptions " Better , Then the computer is stupid .

What do I do , Any effective machine learning algorithm must have its inductive preference , Otherwise it will be assumed that the space appears on the training set " equivalent " Confused by your assumptions , And can't produce certain learning results .

that , Is there any general principle to guide algorithm establishment " Correct " What about preferences ? “ Okam razor ” (Occam’s razor) Is a common 、 The most basic principle in Natural Science Research , namely " If there are more than one hypothesis consistent with observation , Choose the simplest .

summary

Away from specific problems , Talk about in an empty way " What learning algorithm is better " meaningless , Because if you think about all the potential problems , Then all learning algorithms are just as good . We should talk about the relative advantages and disadvantages of the algorithm , Specific learning problems must be addressed ; Good learning algorithm in some problems , On other issues, it may not be satisfactory , Whether the inductive preference of the learning algorithm itself matches the problem , It often plays a decisive role .