Welcome to this friendly beginner’s guide to creating a logistic regression model for classification in python!

欢迎使用这个友好的初学者指南，以在python中创建用于分类的逻辑回归模型！

With this guide I want to give you an easy way to complete your first data science project by creating a logistic regression machine learning model used for classification. I will explain all the steps thoroughly to make sure that you know what is going on. We will be working with a very famous and known data set originating from 1936. I have used Google Colab to write the code.

通过本指南，我想通过创建用于分类的逻辑回归机器学习模型 ，为您提供一个简单的方法来完成您的第一个数据科学项目。 我将彻底解释所有步骤， 以确保您了解正在发生的事情 。 我们将使用源自1936年的非常著名的已知数据集进行工作。我使用Google Colab编写了代码。

The article consists of the following steps:

本文包括以下步骤：

• Introduction to the data数据介绍
• Binary- vs multiclass classification (an explanation)二进制-多类分类(说明)
• Importing the data导入数据
• Splitting the data into training and test data将数据分为训练和测试数据
• Scaling the training and test data扩展训练和测试数据
• Building and training the logistic regression for classification model建立和训练分类模型的逻辑回归
• Evaluating the results评估结果

• Conclusion and bye byes

  结论和再见

I will be showing all the code, and also provide a strong explanation at each step. Sounds good?- Lets do it!

我将展示所有代码，并在每个步骤中都提供有力的解释。 听起来不错？-让我们开始吧！

数据介绍 (Introduction to the data)

The data can be found here. You will be directed to at Kaggle site (see photo below. I have highlighted the download button in yellow). After downloading the file I have renamed it to dataset_iris.csv.

数据可以在这里找到。 您将被带到Kaggle网站(见下图。我以黄色突出显示了下载按钮)。 下载文件后，我将其重命名为dataset_iris.csv 。

The dataset consists of 150 entries of data about iris plants. There are exactly 50 of each type of iris described in the data. The data points for each flower are:

该数据集包含有关虹膜植物的150个数据条目。 数据中描述的每种类型的虹膜正好有50种。 每朵花的数据点是：

1. SepalLengthCmSepalLengthCm
2. SepalWidthCmSepalWidthCm
3. PetalLengthCm花瓣长度
4. PetalWidthCm花瓣宽度
5. Species种类

In total we have 4 variables (1 to 4) that have an influence on which species of plant a given plant might be. In machine learning terms these variables are called features. The final variable (5) is called a label, and it is the label/species that we would like to predict.

总共我们有4个变量(1到4)，这些变量会影响给定植物可能属于哪种植物。 用机器学习术语，这些变量称为功能。 最终变量(5)称为标签，它是我们希望预测的一个或多个标签 。

These are the 20 first rows of the data for you to familiarize with.

这些是您要熟悉的数据的前20行。

二进制与多类分类 (Binary- vs multiclass classification)

For this project you will be creating a logistic regression model for classification. To make the logistic regression model work as a classification model we have to apply a small trick (don’t worry it won’t be difficult). Classification problems can be divided into two categories, binary- and multiclass. The first is a classification problem where the outcome is limited to two different classes. This could be yes or no classifications. Will the customer buy the product or not? Whereas the latter is a classification problem where the outcome can be more than 2 classes, as in the case of this project where we have 3 different classes of plants.

对于此项目，您将创建用于分类的逻辑回归模型。 为了使逻辑回归模型可以用作分类模型，我们必须应用一个小技巧(不要担心这不会很困难)。 分类问题可以分为两类，二类和多类。 第一个是分类问题，其中结果仅限于两个不同的类别。 这可以是是或否的分类。 客户是否会购买产品？ 后者是一个分类问题，其结果可能超过两个类别，例如在该项目中，我们有3种不同类别的植物。

Often logistic regression is not used for classification problems, and there are other models that can naturally do classifications. However, the logistic model is very common and a good model to get familiar with, which is why I have chosen to implement it here anyway. And it does indeed work, we just have to provide the model with a small power up.

通常，逻辑回归不用于分类问题，还有其他一些自然可以进行分类的模型。 但是，逻辑模型非常普遍，并且是一个很好的熟悉模型，因此无论如何我都选择在这里实现它。 它确实有效，我们只需要为模型提供少量电源即可。

Alright, enough talking already, let’s code!Each code snippet will be followed with an explanation of what just happened.

好了，已经说够了，让我们编写代码！每个代码段后面都将出现一个解释。

导入数据 (Importing the data)

import pandas as pddataset = pd.read_csv('dataset_iris.csv')dataset.head()

• We need the pandas library to get access to the data and we import it to a variable called pd.

  我们需要pandas库来访问数据，并将其导入到名为pd的变量中。

• pd (pandas) has a function called read_csv() which allows us to read data from a comma separated file, and it returns a DataFrame which we save to a variable that we call dataset.

  pd (pandas)具有一个称为read_csv()的函数，该函数使我们能够从逗号分隔的文件中读取数据，并且它返回一个DataFrame，我们将其保存到一个称为数据集的变量中。

• A pandas DataFrame has a function called head() which displays the first 5 entries in the DataFrame. (See output photo).

  熊猫DataFrame具有一个名为head()的函数，该函数显示DataFrame中的前5个条目。 (请参见输出照片)。

Now we have all the data saved in a DataFrame format next we will divide the data into features and labels (remember the features are the independent variables that influence the dependent variable called the label).

现在我们将所有数据保存为DataFrame格式，接下来将数据分为要素和标签(请记住，要素是会影响称为标签的因变量的自变量)。

x = dataset.iloc[:, 1:-1].valuesy = dataset.iloc[:, -1].valuesprint(x[:10])print(y[:10])

• We can access different rows and columns of a DataFrame by using iloc followed by the index of the rows and columns we want, and finally appending .values to get the actual values. We save this to variables x and y.

  我们可以使用iloc以及所需的行和列的索引，最后附加.values以获得实际值，从而访问DataFrame的不同行和列。 我们将其保存到变量x和y。

• x now contains 150 rows of features. We see the 10 first rows in the output photo.x现在包含150行要素。 我们在输出照片中看到前10行。
• y now contains the 150 corresponding species/labels. We see the first 10 labels in the output photo.y现在包含150个相应的种类/标签。 我们会在输出照片中看到前10个标签。

We have two more steps to complete before we are ready to create and train our model, let’s do those now!

在准备创建和训练模型之前，我们还有两个步骤需要完成，现在就开始做吧！

将数据分为训练和测试数据 (Splitting the data into training and test data)

from sklearn.model_selection import train_test_splitx_train, x_test, y_train, y_test = train_test_split(x, y, test_size = 0.2, random_state = 0)

• From the scikit-learn library (called sklearn. You will meet sklearn many times!) and model_selection we import train_test_split.

  从scikit-learn库(称为sklearn 。您将多次遇到sklearn ！)和model_selection，我们导入train_test_split。

• train_test_split allows us to split our data intro training data and test data. It takes 4 parameters: the features (x), the labels (y), test_size (what fraction of the data should become test data) and finally random_state (put any number here, but use 0 to get same results as me).

  train_test_split允许我们拆分数据入门培训数据和测试数据。 它包含4个参数： 功能 (x)， 标签 (y)， test_size (数据的小部分应成为测试数据)，最后是random_state (在此处输入任何数字，但使用0可获得与我相同的结果)。

Why do we split the data?When we train a model with data it will get familiar with that data and be able to make perfect predictions when it is introduced to the same data again. It won’t be a lot of fun to only use the model on data it has already seen as we do not want to predict things we already know! Therefore, we split the data and first introduce the test data to the model when it is time to test our models performance.

我们为什么要拆分数据？ 当我们使用数据训练模型时，它将熟悉该数据，并在再次将其引入相同数据时能够做出完美的预测。 仅对已经看到的数据使用该模型并不会带来很多乐趣，因为我们不想预测我们已经知道的事情！ 因此，我们拆分数据，并在需要测试模型性能时首先将测试数据引入模型。

Now to the final step before making the model…

现在到制作模型的最后一步...

缩放数据 (Scaling the data)

We need to scale the data through a method called standardization. (don’t worry about how it works right now, that is a topic for another day). Standardization will scale the values of the features so that almost all of them are in the range of -1 to 1. Some machine learning models are sensitive to whether or not data has been scaled, and logistic regression is one such model. As an example: If we do not scale the data the model might consider 2000m larger than 3km. Scaling will help us get rid of this problem.

我们需要通过一种称为标准化的方法来缩放数据。 (不要担心它现在如何工作，这是另一天的话题)。 标准化将缩放功能部件的值，以便几乎所有功能部件的值都在-1到1的范围内。某些机器学习模型对是否已缩放数据很敏感，而逻辑回归就是这样一种模型。 例如：如果我们不缩放数据，则模型可能会考虑比3km大2000m。 扩展将帮助我们摆脱这个问题。

from sklearn.preprocessing import StandardScalersc = StandardScaler()x_train = sc.fit_transform(x_train)x_test = sc.transform(x_test)

• We import the StandardScaler class from sklearn.preprocessing.

  我们从sklearn.preprocessing导入StandardScaler类。

• We create an instance of StandardScaler and call it sc.

  我们创建StandardScaler的实例，并将其称为sc。

• We call the fit_transform() method from sc, which both fits the scaler to the data and simultaneously scales the training data returning it to the x_train variable.

  我们从sc调用fit_transform()方法， 这既使缩放器适合数据，又同时缩放训练数据并将其返回给x_train变量。

• Lastly we also scale the test data. At this point the scaler is already fitted to the training data and should therefore not be fitted again. We simply call the transform() method which returns the transformed test data.

  最后，我们还扩展了测试数据。 此时，定标器已经安装到训练数据中，因此不应再次安装。 我们只需要调用transform()方法即可返回转换后的测试数据。

• In the output photo after the print statement we can see the first 10 entries of the scaled data. Indeed we can see that the values are mostly between -1 and 1在打印语句后的输出照片中，我们可以看到缩放数据的前10个条目。 实际上，我们可以看到这些值大多在-1和1之间

And now to the step that we have all been waiting for!

现在，我们一直在等待这一步骤！

创建和训练逻辑回归模型 (Creating and training the logistic regression model)

We are now ready to create the logistic regression model for a multiclass problem. Up until now we have: imported the data, split it into training and testing data and lastly scaled the data to make it more suitable for our machine learning model.

现在，我们准备为多类问题创建逻辑回归模型。 到目前为止，我们已经：导入数据，将其分为训练和测试数据，最后缩放数据以使其更适合我们的机器学习模型。

from sklearn.linear_model import LogisticRegressionclassifier = LogisticRegression(multi_class='ovr', random_state = 0)classifier.fit(x_train, y_train)

• We import LogisticRegression from sklearn.linear_model

  我们从sklearn.linear_model导入LogisticRegression

• We create an instance of the LogisticRegression class called classifier by calling LogisticRegression(), as parameters we input multi_class=’ovr’ and random_state=0.

  我们通过调用LogisticRegression()创建名为classifier的LogisticRegression类的实例，作为参数，我们输入multi_class =' ovr '和random_state = 0 。

• multi_class=’ovr’ is the trick I mentioned previously which makes the logistic regression model work in a multiclass scenario. If our problem had simply been binary we would have left this parameter out. Random_state needs any number as input, you can enter 0 to get the same results as me.

  multi_class ='ovr'是我之前提到的技巧，它使逻辑回归模型在多类场景中工作。 如果我们的问题只是二进制问题，我们将忽略此参数。 Random_state需要任何数字作为输入，您可以输入0以获得与我相同的结果。

Wow! That was actually it.

哇！ 就是这样。

Now lets see how well our model performs.

现在让我们看看我们的模型表现如何。

预测结果 (Predicting results)

from sklearn.metrics import confusion_matrix, accuracy_scorecm = confusion_matrix(predictions, y_test)print(cm)accuracy_score(predictions, y_test)

• We import confusion_matrix and accuracy_score. They are kind of self explanatory, however, I will comment a bit on them.

  我们导入confusion_matrix和precision_score 。 它们有点自我解释，但是，我将对它们进行一些评论。

• The confusion_matrix takes two parameters: Our predicted values and the expected values, in this case predictions and y_test. The same applies for the accuracy_score.

  confusion_matrix采用两个参数：我们的预测值和期望值，在这种情况下为预测和y_test。 这同样适用于accuracy_score。

• We print the confusion_matrix called cm

  我们打印出称为cm的confusion_matrix

• The accuracy_score automatically prints out its return value

  precision_score自动打印出其返回值

结论 (Conclusion)

We can see that our model has an accuracy score of 0.9, which means it is correct 90% of the time!

我们可以看到我们的模型的准确度得分是0.9，这意味着90％的时间是正确的！

Congratulations you have now completed a machine learning and data science project!

恭喜，您现在已经完成了机器学习和数据科学项目！

现在要做什么？ (What to do now?)

There are many ways to improve this project! I wanted to keep this guide as simple as possible for everyone to be able to participate. Because of this I have cut some corners and skipped some of the best practices in data science. Normally you would do a lot more data analysis before training the model, like visualizing the data, and also do more work to validate the accuracy of the model. I recommend that you try to complete a project with binary classification as well, so that you master both. Try to do a Google search for linearRegression(multi_class=’ovr’) to completely understand what is going on in this step. Also logistic regression might not be the best machine learning model for this project, I simply chose it because it is so well known and I consider it important to learn. You could try implementing other models in this project and maybe you will achieve better results.

有很多方法可以改善这个项目！ 我希望本指南尽可能地简化，以使每个人都能参与。 因此，我偷偷摸摸走了一些数据科学的最佳实践。 通常，在训练模型之前，您会做很多数据分析，例如可视化数据，并且还要做更多工作来验证模型的准确性。 我建议您也尝试使用二进制分类来完成一个项目，这样您就可以同时掌握两者。 尝试在Google中搜索linearRegression(multi_class ='ovr')，以完全了解此步骤中发生的情况。 另外，逻辑回归可能不是该项目的最佳机器学习模型，我之所以选择它是因为它众所周知，我认为学习很重要。 您可以尝试在该项目中实现其他模型，也许会获得更好的结果。

To get better at AI, machine learning and data science I recommend that you keep practicing. This could include reading other guides here on medium, taking a course on udemy or maybe applying to a boot camp, there are many possibilities! If you are interested in avoiding my biggest mistake for yourself, have a look at my article called My biggest mistake learning machine learning.

为了更好地掌握AI，机器学习和数据科学，我建议您继续练习。 这可能包括在媒体上阅读其他指南，参加关于udemy的课程或申请新兵训练营，这里有很多可能性！ 如果您有兴趣避免自己遇到最大的错误，请参阅我的文章《 我最大的错误学习机器学习》。

I hope this article has been helpful to you! If you have any questions or comments feel free to reach out to me in the response section below.

希望本文对您有所帮助！ 如果您有任何疑问或意见，请随时在下面的回复部分中与我联系。

Keep learning! — Jacob Toftgaard Rasmussen

保持学习！ —雅各布·托夫加德·拉斯穆森