# 机器学习入门

## 1、机器学习是什么

## 2.Classification workflow

### 2.1、Overview

![img](https://gblobscdn.gitbook.com/assets%2F-MW9qnt5mI5Jcz-Mufx8%2F-MWCcUfTaZBSjR9wy8iS%2F-MWCtt9PBLVJzsXXS7PB%2Fimage.png?alt=media\&token=e6b8a9c5-a3e6-48c4-9eb4-82ef694a8826)

### 2.2、Import Data(导入数据)

用到的matlab**函式**

```
readtable("M.txt");
plot();
axis equal;    %使横纵比一样
```

M.txt类似如下:

![img](https://gblobscdn.gitbook.com/assets%2F-MW9qnt5mI5Jcz-Mufx8%2F-MW9qqp3ZbOQzVBLR-40%2F-MWA-GZLZ2E1fSJBVFuO%2Fimage.png?alt=media\&token=fb07cee1-ee93-4856-82c3-ecc024a39382)

### 2.3、Process Data(处理数据)

### 2.4、Extract Features(特征提取)

![img](https://gblobscdn.gitbook.com/assets%2F-MW9qnt5mI5Jcz-Mufx8%2F-MW9qqp3ZbOQzVBLR-40%2F-MW9qwVW2rm0DXLBcrXh%2Fimage.png?alt=media\&token=4880cb3c-2173-4bfd-b1fd-80e5efae1b76)

通过计算不同的**特征**来辨别字母

* 例如J和M，J会窄一点，而M近似于square，所以可以通过Y轴范围和X轴范围之比来确定
* 相比于J和M，V会写的快一点，所以可以把duration当作特征值来辨别

在这里我们计算两个特征

* letter的**长宽比**：AspectRatio
* letter的书写时间：Duration

```
AspectRatio = range(letter.Y)/range(letter.X);
Duration = letter.Time(end) - letter.Time(1);

%range(x),  x is an array
%means  max(x)-min(x)
```

![img](https://gblobscdn.gitbook.com/assets%2F-MW9qnt5mI5Jcz-Mufx8%2F-MW9qqp3ZbOQzVBLR-40%2F-MW9xb07Md3fVLBSv6mc%2Fimage.png?alt=media\&token=8a3c4934-fec4-454a-9ee8-c084a4ccfe09)

featuredata.mat:

![img](https://gblobscdn.gitbook.com/assets%2F-MW9qnt5mI5Jcz-Mufx8%2F-MW9qqp3ZbOQzVBLR-40%2F-MW9xuOfF9U_hunA6ppe%2Fimage.png?alt=media\&token=1657ac61-063c-4fb4-943b-9c62a79e329b)

通过scatter和gscatter两个函式来visualization（散点图）

```
load featuredata.mat
features

scatter(features.AspectRatio,features.Duration);

%gscatter 是可以通过第三个parameter来展现不同的颜色
gscatter(features.AspectRatio,features.Duration,features.Character)
```

scatter:

![img](https://gblobscdn.gitbook.com/assets%2F-MW9qnt5mI5Jcz-Mufx8%2F-MW9qqp3ZbOQzVBLR-40%2F-MW9yaOH9rR8dZp8brQB%2Fimage.png?alt=media\&token=3354a409-0bfd-46db-bd28-d1635f80fc58)

gscatter:

![img](https://gblobscdn.gitbook.com/assets%2F-MW9qnt5mI5Jcz-Mufx8%2F-MW9qqp3ZbOQzVBLR-40%2F-MW9yok3MR7Pc4rX1YR2%2Fimage.png?alt=media\&token=3bade2e4-3d5f-4342-b4c5-90895433565e)

### 2.5、Build a Model

![img](https://gblobscdn.gitbook.com/assets%2F-MW9qnt5mI5Jcz-Mufx8%2F-MW9qqp3ZbOQzVBLR-40%2F-MWA-gyVr-8v5uSShMuQ%2Fimage.png?alt=media\&token=e56f2131-5fad-466c-bcbd-9638fbc91678)

![img](https://gblobscdn.gitbook.com/assets%2F-MW9qnt5mI5Jcz-Mufx8%2F-MW9qqp3ZbOQzVBLR-40%2F-MWA034-jqpJp2V1Fh5w%2Fimage.png?alt=media\&token=00bfcb2a-58f3-40ab-aeb8-ae4a0d06dfe7)

在本例中，我们应用**KNN模型**，并用matlab里面自带的KNN分类器

* 关于KNN（[https://zhuanlan.zhihu.com/p/25994179）](https://zhuanlan.zhihu.com/p/25994179%EF%BC%89)

加载table

```
load featuredata.mat
features    %features包括有两个维度的特征以及相应的            %label，这里的label是Character
```

![img](https://gblobscdn.gitbook.com/assets%2F-MW9qnt5mI5Jcz-Mufx8%2F-MWCcUfTaZBSjR9wy8iS%2F-MWCcYbKNxrX4N2IKmhr%2Fimage.png?alt=media\&token=c6dce066-dbc0-4a85-9364-04b4e449f96c)

用matlab的函式构造模型并预测

```
knnmodel = fitcknn(features, "Character");
newdata = [4,1.2];    %parameter1是AspectRatio ...
predicted = predict(knnmodel, newdata);
```

调整K邻近的K参数

![img](https://gblobscdn.gitbook.com/assets%2F-MW9qnt5mI5Jcz-Mufx8%2F-MWCcUfTaZBSjR9wy8iS%2F-MWCctioyvvUi59HEe3S%2Fimage.png?alt=media\&token=96d5afb6-cc41-45b8-b3d3-dd86955f4dc1)

预测值

![img](https://gblobscdn.gitbook.com/assets%2F-MW9qnt5mI5Jcz-Mufx8%2F-MWCcUfTaZBSjR9wy8iS%2F-MWCd4zgPkAPeoMywzVr%2Fimage.png?alt=media\&token=81189e2a-b819-4475-b562-9db5c730f84a)

### 2.6、Evaluate Model（评价模型）

在本例中，我们有三种evaluate的方式

* accuracy
* misclassrate
* confusion matrix（混淆矩阵）

我们通过training data得到**model**

再通过test data来evaluate 我们的model好坏

这里的**混淆矩阵**用到了matlab的函式

```
%ytrue is a vector of the known classes
%ypred is a vector of the predicted classes
confusionchart(ytrue, ypred);
```

**confusion matrix**

![img](https://gblobscdn.gitbook.com/assets%2F-MW9qnt5mI5Jcz-Mufx8%2F-MWCcUfTaZBSjR9wy8iS%2F-MWClKWP0ROe9qSFYcLb%2Fimage.png?alt=media\&token=9fc4d0e8-d6a7-462e-a48a-28212e44af56)

![img](https://gblobscdn.gitbook.com/assets%2F-MW9qnt5mI5Jcz-Mufx8%2F-MWCcUfTaZBSjR9wy8iS%2F-MWClPVlCMpkuk73TlGv%2Fimage.png?alt=media\&token=3b3d9d8b-ee5a-4968-88ba-c4e50443ff51)

### 2.7、Review

这次我们用的是2个特征，来辨别13个字母

我们把所有流程回顾

![img](https://gblobscdn.gitbook.com/assets%2F-MW9qnt5mI5Jcz-Mufx8%2F-MWCcUfTaZBSjR9wy8iS%2F-MWCsh8fLyWsxEjzHAXp%2Fimage.png?alt=media\&token=9f99cd0a-39b8-4921-8603-bbcf4958c353)

```
load featuredata13letters.mat
features
testdata

gscatter(features.AspectRatio,features.Duration,features.Character)
xlim([0 10])

knnmodel = fitcknn(features,"Character","NumNeighbors",5);
predictions = predict(knnmodel,testdata);

misclass = sum(predictions ~= testdata.Character)/numel(predictions)
confusionchart(testdata.Character,predictions);
```

gscatter:

![img](https://gblobscdn.gitbook.com/assets%2F-MW9qnt5mI5Jcz-Mufx8%2F-MWCcUfTaZBSjR9wy8iS%2F-MWCtH05fvcnZO76z-V3%2Fimage.png?alt=media\&token=ab135178-b868-4f62-9952-cdc436aaf817)

confusion matrix

![img](https://gblobscdn.gitbook.com/assets%2F-MW9qnt5mI5Jcz-Mufx8%2F-MWCcUfTaZBSjR9wy8iS%2F-MWCtRXvpMNJlwu5WvzX%2Fimage.png?alt=media\&token=2560f014-f3eb-4d84-9eee-c125bff3b338)

## 3.Importing and Preprocessing Data

### 3.2、Creating Datastores

用到了三个matlab函式

* letterds = datastore("file")
* data = read(letterds)
* data = readall(letterds)

```
%读取xxxx_M_xxxx.txt的文件
letterds = datastore("*_M_*.txt");

%读取letterds的第一个文件
data = read(letterds);

%读取letterds的第二个文件
data = read(letterds);

%读取letterds的所有文件
data = readall(letterds);
```

### 3.3、Adding a Data Transformation

在**读datastore**之前先将其**transformation**

* 例如normalization

```
%load files to datastore
letters = datastore("*_M_*.txt");

preprocd = transform(letters, @scale);
%注意这个function 要在transform后面
function data = scale(data)
    data.Time = (data.Time - data.Time(1))/1000;
    data.X = 1.5*data.X;
    %omitnan表示计算mean时ignore掉NAN
    data.X = data.X - mean(data.X,"omitnan");
    data.Y = data.Y - mean(data.Y,"omitnan");
end
```

### supplement

normalization:

* \[0\~1] 
* \[-1\~1]

![img](https://gblobscdn.gitbook.com/assets%2F-MW9qnt5mI5Jcz-Mufx8%2F-MWDehM9Xj2oB6UNOBD5%2F-MWDeitCbxtNHMFzNIP3%2Fimage.png?alt=media\&token=b5ec8e94-f83d-4adb-bb76-c0b224d7d49d)

## 4.Engineering Features(工程特征)

### 4.1、Types of Signals

### 4.2、Calculating Summary Statistics

#### Statistical Functions

#### Measures of Central Tendency

|                              Function                             |               Description               |
| :---------------------------------------------------------------: | :-------------------------------------: |
|   [`mean`](https://www.mathworks.com/help/matlab/ref/mean.html)   |             Arithmetic mean             |
| [`median`](https://www.mathworks.com/help/matlab/ref/median.html) |          Median (middle) value          |
|   [`mode`](https://www.mathworks.com/help/matlab/ref/mode.html)   |           Most frequent value           |
|  [`trimmean`](https://www.mathworks.com/help/stats/trimmean.html) | Trimmed mean (mean, excluding outliers) |
|   [`geomean`](https://www.mathworks.com/help/stats/geomean.html)  |              Geometric mean             |
|  [`harmean`](https://www.mathworks.com/help/stats/harmmean.html)  |              Harmonic mean              |

#### Measures of Spread

|                                    Function                                    |                         Description                         |
| :----------------------------------------------------------------------------: | :---------------------------------------------------------: |
|           [`range`](https://www.mathworks.com/help/stats/range.html)           |             Range of values (largest – smallest)            |
|           [`std`](https://www.mathworks.com/help/matlab/ref/std.html)          |                      Standard deviation                     |
|           [`var`](https://www.mathworks.com/help/matlab/ref/var.html)          |                           Variance                          |
|             [`mad`](https://www.mathworks.com/help/stats/mad.html)             |                   Mean absolute deviation                   |
| [`iqr`](https://www.mathworks.com/help/stats/prob.normaldistribution.iqr.html) | Interquartile range (75th percentile minus 25th percentile) |

#### Measures of Shape

|                             Function                             |            Description            |
| :--------------------------------------------------------------: | :-------------------------------: |
| [`skewness`](https://www.mathworks.com/help/stats/skewness.html) |  Skewness (third central moment)  |
| [`kurtosis`](https://www.mathworks.com/help/stats/kurtosis.html) |  Kurtosis (fourth central moment) |
|   [`moment`](https://www.mathworks.com/help/stats/moment.html)   | Central moment of arbitrary order |

* the median is less sensitive to outliers than the mean
* the MAD(mean absolute deviation) is:

  ![img](https://gblobscdn.gitbook.com/assets%2F-MW9qnt5mI5Jcz-Mufx8%2F-MWHgHCgX7cfZgVqUUWS%2F-MWHiPEF2SZlmCrxfh0H%2Fimage.png?alt=media\&token=d656ab8f-f6c3-4032-91b4-3d05fecd9129)

## 4.3、Finding Peaks

使用matlab的函式来寻找局部maxima和minima

* \[idx, prom] = islocalmin(X, options)
* idx is logical array
* prom是取到局部最大（最小）的具体值

![img](https://gblobscdn.gitbook.com/assets%2F-MW9qnt5mI5Jcz-Mufx8%2F-MWHgHCgX7cfZgVqUUWS%2F-MWHnHfByGF6uCmNtie8%2Fimage.png?alt=media\&token=78feb647-b133-4da7-a0a8-8841a9a6698d)

通过threshold来去除噪点，因为判断最大最小值，只要看这个点的两边就ok

例如：![img](https://gblobscdn.gitbook.com/assets%2F-MW9qnt5mI5Jcz-Mufx8%2F-MWHgHCgX7cfZgVqUUWS%2F-MWHnyOyENt9Ah2KPIdv%2Fimage.png?alt=media\&token=0eb22cfa-60df-48e5-989d-5c3ec63ca7c8)

## 4.4、Computing Derivatives(计算导数)

用到matlab的函式

* diff(X)    --->  \[x2-x1,x3-x2,....]
* X = standardizeMissing(X, value)   ---> 将value转成NAN

在本例中，我们通过字母的X轴和Time来计算在书写速率velocity

并用standardizeMissing来去除计算出inf的影响

## 4.5、Calculating Correlations

用到matlab的函式：

* corr(X,Y) --->求X,Y的相关系数

  ![img](https://gblobscdn.gitbook.com/assets%2F-MW9qnt5mI5Jcz-Mufx8%2F-MWKYl80ly6gFWv6ZqaI%2F-MWKYo6dpkE5VTxAryd-%2Fimage.png?alt=media\&token=4b82c09f-1ab5-4463-9f1b-de539afc5c58)
* 衡量X,Y之间的**线性**相关度

![image-20210322001408361](C:%5CUsers%5Chasee%5CAppData%5CRoaming%5CTypora%5Ctypora-user-images%5Cimage-20210322001408361.png)

计算X和Y的相关系数，得出NAN，是因为有MISSING DATA

接下来用...来去掉missing data

```
corr(x,y,"Rows","complete")
```

在本例中v2.X和v2.Y为22\*1的列向量

![img](https://gblobscdn.gitbook.com/assets%2F-MW9qnt5mI5Jcz-Mufx8%2F-MWKYl80ly6gFWv6ZqaI%2F-MWKZh7CvU3wMO_FF7cC%2Fimage.png?alt=media\&token=7d234232-88d6-4ee9-be66-980e3ba78126)

```
M = [v2.X(1:11),v2.Y(1:11),v2.X(12:22),v2.Y(12:22)];
```

![img](https://gblobscdn.gitbook.com/assets%2F-MW9qnt5mI5Jcz-Mufx8%2F-MWKYl80ly6gFWv6ZqaI%2F-MWKZvebxQFLhM2VQREY%2Fimage.png?alt=media\&token=8c2c5ee3-59be-4fc7-908d-402666861877)

```
Cmat = corr(M,"Rows","complete");
```

Cmat:

![img](https://gblobscdn.gitbook.com/assets%2F-MW9qnt5mI5Jcz-Mufx8%2F-MWKYl80ly6gFWv6ZqaI%2F-MWK_50r_GwRmUJPrrJz%2Fimage.png?alt=media\&token=8c8102dd-f485-4119-b463-f349c909fc58)

* Cmat(i,j)表示第i列和第j列的相关系数
* 可以看到Cmat是对称矩阵