Introduction

介绍

For this article, I’d like to introduce you to KNN with a practical example.

对于本文，我想通过一个实际的例子向您介绍KNN。

I will consider one of my project that you can find in my GitHub profile. For this project, I used a dataset from Kaggle.

我将考虑可以在我的GitHub个人资料中找到的我的项目之一。 对于这个项目，我使用了Kaggle的数据集。

The dataset is the result of a chemical analysis of wines grown in the same region in Italy but derived from three different cultivars organized in three classes. The analysis was done by considering the quantities of 13 constituents found in each of the three types of wines.

该数据集是对意大利同一地区种植的葡萄酒进行化学分析的结果，这些葡萄酒来自三个不同类别的三个品种。 通过考虑三种葡萄酒中每种葡萄酒中13种成分的数量来进行分析。

This article will be structured in three-part. In the first part, I will make a theoretical description of KNN, then I will focus on the part about exploratory data analysis in order to show you the insights that I found and at the end, I will show you the code that I used to prepare and evaluate the machine learning model.

本文将分为三部分。 在第一部分中，我将对KNN进行理论上的描述，然后，我将重点介绍探索性数据分析这一部分，以便向您展示我发现的见解，最后，我将向您展示我曾经使用过的代码准备和评估机器学习模型。

Part I: What is KNN and how it works mathematically?

第一部分：什么是KNN及其在数学上的作用？

The k-nearest neighbour algorithm is not a complex algorithm. The approach of KNN to predict and classify data consists of looking through the training data and finds the k training points that are closest to the new point. Then it assigns to the new data the class label of the nearest training data.

k最近邻居算法不是复杂的算法。 KNN预测和分类数据的方法包括浏览训练数据并找到最接近新点的k个训练点。 然后，它将新的训练数据的类别标签分配给新数据。

But how KNN works? To answer this question we have to refer to the formula of the euclidian distance between two points. Suppose you have to compute the distance between two points A(5,7) and B(1,4) in a Cartesian plane. The formula that you will apply is very simple:

但是KNN是如何工作的？ 要回答这个问题，我们必须参考两点之间的欧几里得距离的公式。 假设您必须计算笛卡尔平面中两个点A(5,7)和B(1,4)之间的距离。 您将应用的公式非常简单：

Okay, but how can we apply that in machine learning? Imagine to be a bookseller and you want to classify a new book called Ubick of Philip K. Dick with 240 pages which cost 14 euro. As you can see below there are 5 possible classes where to put our new book.

好的，但是我们如何将其应用到机器学习中呢？ 想象成为一个书商，您想对一本名为Philip K. Dick的Ubick的新书进行分类，共有240页，售价14欧元。 如您在下面看到的，有5种可能的类别可用于放置我们的新书。

To know which is the best class for Ubick we can use the euclidian formula in order to compute the distance with each observation in the dataset.

要知道哪个是Ubick的最佳分类，我们可以使用欧几里得公式来计算数据集中每个观测值的距离。

Formula:

式：

output:

输出：

As you can see above the nearest class for Ubick is class C.

如您所见，Ubick最近的课程是C类 。

Part II: insights that I found to create the model

第二部分：我发现的创建模型的见解

Before to start to speak about the algorithm, that I used to create my model and predict the varieties of wine, let me show you briefly the main insights that I found.

在开始谈论算法之前，我曾用它来创建模型并预测葡萄酒的种类，然后让我简要地向您展示我发现的主要见解。

In the following heatmap, there are correlations between the different features. This is very useful to have a first look at the situation of our dataset and knowing if it is possible to apply a classification algorithm.

在下面的热图中，不同功能之间存在关联。 首先了解一下数据集的情况，并了解是否有可能应用分类算法，这非常有用。

The heatmap is great for a first look but that is not enough. I’d like also to know if there are some elements whose absolute sum of correlations is low in order to delete them before to train the machine learning model. So, I construct a histogram as you can see below.

该热图乍一看很棒，但这还不够。 我还想知道是否存在某些元素的相关绝对和很低，以便在训练机器学习模型之前将其删除。 因此，如下图所示，我构建了一个直方图。

You can see that there are three elements with low total absolute correlation. The elements are ash, magnesium and the color_intensity.

您会看到三个绝对绝对相关性较低的元素。 元素是灰，镁和color_intensity。

Thanks to these observations now we are sure that there is the possibility to apply a KNN algorithm to create a predictive model.

现在，由于这些观察，我们确信可以应用KNN算法创建预测模型。

Part III: use scikit-learn to make predictions

第三部分：使用scikit-learn进行预测

In this part, we will see how to prepare the model and evaluate it thanks to scikit-learn.

在这一部分中，我们将借助scikit-learn了解如何准备模型并进行评估。

Below you can observe that I split the model into two parts: 80% for training and 20% for testing. I chose this proportion because the data set is not big.

在下面，您可以看到我将模型分为两个部分：80％用于训练，20％用于测试。 我选择此比例是因为数据集不大。

# split data to train and test
y = df['class']
X = input_data.drop(columns=['ash','magnesium', 'color_intensity'])X_train, X_test, y_train, y_test = train_test_split(X, y, test_size = 0.20, random_state=0)# to be sure that the data was split rightly (80% for train data and 20% for test data)print("X_train shape: {}".format(X_train.shape))
print("y_train shape: {}".format(y_train.shape))print("X_test shape: {}".format(X_test.shape))
print("y_test shape: {}".format(y_test.shape))

out:

出：

X_train shape: (141, 10)
y_train shape: (141,)X_test shape: (36, 10)
y_test shape: (36,)

You have to know that all machine learning models in scikit-learn are implemented in their own classes. For example, the k-nearest neighbors classification algorithm is implemented in the KNeighborsClassifier class.

您必须知道scikit-learn中的所有机器学习模型都是在各自的类中实现的。 例如，在KNeighborsClassifier类中实现了k最近邻居分类算法。

The first step is to instantiate the class into an object that I called cli as you can see below. The object contains the algorithm that I will use to build the model from the training data and make predictions on new data points. It contains also the information that the algorithm has extracted from the training data.

第一步是将类实例化为一个我称为cli的对象，如下所示。 该对象包含用于从训练数据构建模型并对新数据点进行预测的算法。 它还包含算法已从训练数据中提取的信息。

Finally, to build the model on the training set, we call the fit method of the cli object.

最后，要在训练集上构建模型，我们调用cli对象的fit方法 。

from sklearn.neighbors import KNeighborsClassifiercli = KNeighborsClassifier(n_neighbors=1)
cli.fit(X_train, y_train)

out:

出：

KNeighborsClassifier(algorithm='auto', leaf_size=30, metric='minkowski',metric_params=None, n_jobs=None, n_neighbors=1, p=2,weights='uniform')

In the output of the fit method, you can see the parameters used in creating the model.

在fit方法的输出中，您可以看到用于创建模型的参数。

Now, it is time to evaluate the model. Below, the first output shows us that the model predict the 89% of the test data. Instead the second output give us a complete overview of the accuracy for each class.

现在，该评估模型了。 下面的第一个输出向我们展示了该模型预测了89％的测试数据。 相反，第二个输出为我们提供了每个类别的准确性的完整概述。

y_pred = cli.predict(X_test)
print("Test set score: {:.2f}".format(cli.score(X_test, y_test))) # below the values of the model 
from sklearn.metrics import classification_report
print("Final result of the model \n {}".format(classification_report(y_test, y_pred)))

out:

出：

Test set score: 0.89

out:

出：

Conclusion

结论

I think that the best way to learn something is by practising. So in my case, I download the dataset from Kaggle which is one of the best places where to find a good dataset on which you can apply your machine learning algorithms and learn how they work.

我认为最好的学习方法是练习。 因此，就我而言，我是从Kaggle下载数据集的，这是找到良好数据集的最佳位置之一，您可以在该数据集上应用机器学习算法并了解它们的工作方式。

Thanks for reading this. There are some other ways you can keep in touch with me and follow my work:

感谢您阅读本文。 您可以通过其他方法与我保持联系并关注我的工作：

• Subscribe to my newsletter.

  订阅我的时事通讯。

• You can also get in touch via my Telegram group, Data Science for Beginners.

  您也可以通过我的电报小组“ 面向初学者的数据科学”来联系 。