Machine learning practice: is Mermaid a love movie or an action movie? KNN announces the answer

Author's brief introduction ： The most important part of the whole building is the foundation , The foundation is unstable , The earth trembled and the mountains swayed . And to learn technology, we should lay a solid foundation , Pay attention to me , Take you to firm the foundation of the neighborhood of each plate .
Blog home page ： Qigui's blog
Included column ：《 Statistical learning method 》 The second edition —— Personal notes
From the south to the North , Don't miss it , Miss this article ,“ wonderful ” May miss you la
Triple attack( Three strikes in a row ):Comment,Like and Collect—>Attention

problem ：

It is known that 《 Warwolf 》《 The red sea action 》《 Mission: impossible 6》 Action movie. , and 《 The former 3》《 Chunjiao saves Zhiming 》《 Titanic Number 》 It's romance. . But if ⼀ Dan now has ⼀ A new movie 《 Mermaid 》, Is there any ⼀ There are two ways for the machine to master ⼀ Rules of classification , Automatically classify new movies ？

Visualize the data

Because the data in the above table does not look intuitive , So visualize .
The code is as follows ：

#  Data visualization , Visually observe the classification of movie data 
import matplotlib.pyplot as plt

# matplotlin Chinese display is not supported by default —— adopt rcParams Configuration parameters 
plt.rcParams['font.sans-serif'] = ['SimHei'] #  Step one （ Replace sans-serif typeface ）
plt.rcParams['axes.unicode_minus'] = False   #  Step two （ Solve the problem of negative sign display of coordinate axis negative numbers ）

# 1. get data 
# 2. Basic data processing 
# 3. Data visualization 

x = [5,3,31,59,60,80]		# x The axis represents the number of kisses 
y = [100,95,105,2,3,10]		# y The axis represents the number of fights 
labels = ["《 Warwolf 》","《 The red sea action 》","《 Mission: impossible 6》","《 The former 3》","《 love in a puff 》","《 Titanic 》"]

#  Set scales and labels 
plt.scatter(x,y,s=120)
plt.xlabel(" Number of kisses ")
plt.ylabel(" Number of fights ")
plt.xticks(range(0,150,10))
plt.yticks(range(0,150,10))

#  Add annotation text to each point 
count = 0
for x_i,y_i in list(zip(x,y)):	#  use zip take x,y package , Then turn to the list 
    plt.annotate(f"{
      labels[count]}",xy=(x_i,y_i),xytext=(x_i,y_i))
    count+=1
    
plt.show()

The output result is shown below ：

thus , The data in the table is clear at a glance .
But we don't know 《 Mermaid 》 Which point is closer to this point , How to calculate 《 Mermaid 》 The distance between this point and other points ？—— Two point distance

The distance between two points is also called Euclidean distance ：$d=\sqrt{{\left(x_1-y_1\right)}^2+{\left(x_2-y_2\right)}^2}$
Know the article in detail ：
Concept to method , great 《 Statistical learning method 》—— The third chapter 、k Nearest neighbor method

current demand It's calculation 《 Mermaid 》 Distance from each movie . We may all think , Only look for ⼀ The nearest neighbor to judge 《 Mermaid 》 Isn't it just the type of ？ In fact, this approach is not feasible .
Here's a chestnut ：

In the figure A Classmates and B、D The distance between the two students is equal , These two students are in different districts , Then you can't judge A Which district does the student belong to . in other words , We should find more neighbors , In order to determine its classification more accurately . Compare this movie classification to determine the selection 3 A close neighbor , in other words k=3.

In the last article （ Concept to method , great 《 Statistical learning method 》—— The third chapter 、k Nearest neighbor method ） As detailed in KNN Workflow , Here is a brief memory .

summary KNN Workflow ：

• 1. Calculate the distance between the object to be classified and other objects ;
• 2. Count the nearest K A neighbor ;
• 3. about K The nearest neighbor , Which category do they belong to the most , Where the object to be classified belongs ⼀ class .

There are two implementations ：

1、Numpy Realization

import numpy as np
import pandas as pd
import matplotlib.pyplot as plt

plt.rcParams['font.sans-serif'] = ['SimHei']  #  Step one （ Replace sans-serif typeface ）
plt.rcParams['axes.unicode_minus'] = False  #  Step two （ Solve the problem of negative sign display of coordinate axis negative numbers ）

# 1.  Prepare the data 
#  Training data characteristics and objectives 
#  Test data characteristics and objectives 

# 2.  Calculate the Euclidean distance between unknown data features and training data features 
# 3.  Count the nearest K A neighbor 
# 4.  Make category statistics 

class Myknn(object):
    def __init__(self, train_df, k):
        """ :param train_df:  Training data  :param k:  Number of nearest neighbors  """
        self.train_df = train_df
        self.k = k

	#  Calculate forecast data   And   Training data   distance 
	#  Select minimum distance k It's worth 
	#  Calculation k Of the values   Categories   Proportion 
    def predict(self, test_df):
        """ Prediction function """
        #  Calculate the Euclidean distance —— Use nupy Calculated distance 
        self.train_df[' distance '] = np.sqrt(
            (test_df[' Number of fights '] - self.train_df[' Number of fights ']) ** 2 + (test_df[' Number of kisses '] - self.train_df[' Number of kisses ']) ** 2)
            
        #  Sort by distance   Get the one with the smallest distance   front K Categories of data 
        my_types = self.train_df.sort_values(by=' distance ').iloc[:self.k][' Film type ']
        print(my_types)
        
		#  Yes k Classification of points for statistics , See who accounts for more , The predicted value belongs to which category 
        new_my_type = my_types.value_counts().index[0]
        print(new_my_type)


# 1.  Reading data 
my_df = pd.read_excel(' Movie data .xlsx', sheet_name=0)
# print(my_df)
# 2. Basic data processing 
# 3. Feature Engineering 
# 4. Data visualization （ Omit ）

# 2.  Training data   Characterized features : The number of fights and kisses   Initialize tags ( Category )： Film type 
train_df = my_df.loc[:5, [" Number of fights ", " Number of kisses ", " Film type "]]
# 3.  Forecast data 
test_df = my_df.loc[6, [" Number of fights ", " Number of kisses ", ]]

# 4.  Algorithm implementation —— Instantiate the class 
mk = Myknn(train_df, 3)  # k=3
mk.predict(test_df)

Output results ：

3     love 
4     love 
5     love 
Name:  Film type , dtype: object
 love 

2、Sklearn Realization

sklearn.neighbors.KNeighborsClassifier(n_neighbors=5, weights=‘uniform’, algorithm=‘auto’)

• n_neighbors： By default kneighbors Number of neighbors used by query . Namely k-NN Of k Value , Pick the nearest k A little bit . Number of adjacent points , namely k The number of .（ The default is 5）
• weights： The default is “uniform” It means that each nearest neighbor is assigned the same weight , That is to determine the weight of the nearest neighbor ; Can be specified as “distance” It indicates that the distribution weight is inversely proportional to the distance of the query point ; At the same time, you can also customize the weight .
• algorithm: Nearest neighbor algorithm , Optional {‘auto’, ‘ball_tree’, ‘kd_tree’, ‘brute’}. Optional fast k Nearest neighbor search algorithm , The default parameter is auto, It can be understood that the algorithm decides the appropriate search algorithm by itself ; You can also specify your own search algorithm ball_tree、kd_tree、brute Method to search ,brute It's brute force search , That's linear scanning .

step ：

• 1、 Build feature data and target data
• 2、 structure k A nearest neighbor classifier
• 3、 send ⽤fit Training
• 4、 Forecast data

from sklearn.neighbors import KNeighborsClassifier

# 1. Reading data 
mv_df = pd.read_excel(" Movie data .xlsx",sheet_name=0)

# 2. Build the characteristic data of the training set —— Be careful ： Two dimensional arrays 
x = mv_df.loc[:5," Number of fights ":" Number of kisses "].values

# 3. Build the target data of the training set —— Be careful ： One dimensional array 
y = mv_df.loc[:5," Film type "].values

# 4. Instantiation KNN Algorithm ( classifier )
knn = KNeighborsClassifier(n_neighbors=4)

# 5. Training —— Characteristic data of training set   Target data of training set 
knn.fit(x,y)

# 6. Forecast data 
knn.predict([[5,29]])

Output results ：

[' love ']

therefore , Through training KNN Model , Finally it is concluded that 《 Mermaid 》 It's a love movie .

The learning ：
Bloggers are determined to understand every knowledge point . Why? ？ I've read an article about the learning experience shared by a big man , One of them is very good ：【 Basic knowledge is like the foundation of a building , It determines the height of our technology , And to do something quickly , The precondition must be that the basic ability is excellent ,“ Internal skill ” In place 】. therefore , Since then, I have focused on a solid foundation , And I know , Learning technology can't be learned overnight , Not for half a year 、 A year , Or two years , Even longer ; But I believe , Persistence is never wrong . Last , I want to say ： Focus on the basics , Establish your own knowledge system .

Okay , This learning sharing is over . Mix a —— How many of the above seven movies have you seen ？ Which one do you like best ？ Tell me in the comment area ！ Hey ,, I can't help but hurry to see it before studying .

Knowledge review ：
Concept to method , great 《 Statistical learning method 》—— The third chapter 、k Nearest neighbor method