How to build a decision tree in Python

Decision Tree

Decision trees are an integral part of today's most powerful supervised learning methods. A decision tree is basically a flowchart of a binary tree where each node splits a set of observations based on some characteristic variable.

The goal of a decision tree is to divide data into groups so that every element in a group belongs to the same category. Decision trees can also be used to approximate continuous target variables. In this case, the tree will be split such that each group has the smallest mean squared error.

An important property of decision trees is that they are easily interpretable. You don't need to be familiar with machine learning techniques at all to understand what decision trees are doing. Decision tree diagrams are easy to interpret.

Pros and Cons

The advantages of the decision tree method are:

• The decision tree can generate understandable rules.

• Decision trees perform classification without requiring extensive calculations.

• Decision trees can handle continuous variables and categorical variables.

• Decision trees provide a clear indication of which fields are most important.

The disadvantages of the decision tree method are:

• Decision trees are not suitable for estimation tasks where the goal is to predict continuous attribute values.

• Decision trees are prone to errors in classification problems with many classes and few training samples.

• Training of decision trees can be computationally expensive. The process of generating decision trees is computationally very expensive. At each node, each candidate split field must be sorted to find its best split. In some algorithms, using combinations of fields, the best combination of weights must be searched. Pruning algorithms can also be expensive because many candidate subtrees must be formed and compared.

Python Decision Tree

Python is a general-purpose programming language that provides data scientists with powerful machine learning packages and tools. In this article, we will use scikit-learn, the most famous machine learning package in Python, to build a decision tree model. We will create the model using the "DecisionTreeClassifier" algorithm provided by scikit learn and then visualize the model using the "plot_tree" function.

Step 1: Import package

The main software packages we use to build models are pandas, scikit learn and NumPy. Follow the code to import the required packages in python.

import pandas as pd # 数据处理 import numpy as np # 使用数组 import matplotlib.pyplot as plt # 可视化 from matplotlib import rcParams # 图大小 from termcolor import colored as cl # 文本自定义  from sklearn.tree import DecisionTreeClassifier as dtc # 树算法 from sklearn.model_selection import train_test_split # 拆分数据 from sklearn.metrics import accuracy_score # 模型准确度 from sklearn.tree import plot_tree # 树图  rcParams['figure.figsize'] = (25, 20)

After importing all the packages needed to build our model, it’s time to import the data and do some EDA on it.

Step 2: Import data and EDA

In this step we will use the "Pandas" package provided in python to import and do some EDA on it . We will build our decision tree model on a dataset that is a dataset of medications that are prescribed to patients based on specific criteria. Let’s import the data using python!

Python implementation:

df = pd.read_csv('drug.csv') df.drop('Unnamed: 0', axis = 1, inplace = True)  print(cl(df.head(), attrs = ['bold']))

Output:

   Age Sex      BP Cholesterol  Na_to_K   Drug 0   23   F    HIGH        HIGH   25.355  drugY 1   47   M     LOW        HIGH   13.093  drugC 2   47   M     LOW        HIGH   10.114  drugC 3   28   F  NORMAL        HIGH    7.798  drugX 4   61   F     LOW        HIGH   18.043  drugY

Now we have a clear idea of ​​the dataset. After importing the data, let's use the "info" function to get some basic information about the data. Information provided by this function includes number of entries, index number, column name, non-null value count, attribute type, etc.

Python implementation:

df.info()

Output:

 RangeIndex: 200 entries, 0 to 199 Data columns (total 6 columns):  #   Column       Non-Null Count  Dtype   ---  ------       --------------  -----    0   Age          200 non-null    int64    1   Sex          200 non-null    object   2   BP           200 non-null    object   3   Cholesterol  200 non-null    object   4   Na_to_K      200 non-null    float64  5   Drug         200 non-null    object  dtypes: float64(1), int64(1), object(4) memory usage: 9.5+ KB

Step 3: Data processing

We can See that attributes like Sex, BP and Cholesterol are categorical and object type in nature. The problem is that the decision tree algorithm in scikit-learn inherently does not support X variables (features) being of type "object". Therefore, it is necessary to convert these "object" values ​​into "binary" values. Let us use python to implement

Python implementation:

for i in df.Sex.values:     if i  == 'M':         df.Sex.replace(i, 0, inplace = True)     else:         df.Sex.replace(i, 1, inplace = True)  for i in df.BP.values:     if i == 'LOW':         df.BP.replace(i, 0, inplace = True)     elif i == 'NORMAL':         df.BP.replace(i, 1, inplace = True)     elif i == 'HIGH':         df.BP.replace(i, 2, inplace = True)  for i in df.Cholesterol.values:     if i == 'LOW':         df.Cholesterol.replace(i, 0, inplace = True)     else:         df.Cholesterol.replace(i, 1, inplace = True)  print(cl(df, attrs = ['bold']))

Output:

     Age  Sex  BP  Cholesterol  Na_to_K   Drug 0     23    1   2            1   25.355  drugY 1     47    1   0            1   13.093  drugC 2     47    1   0            1   10.114  drugC 3     28    1   1            1    7.798  drugX 4     61    1   0            1   18.043  drugY ..   ...  ...  ..          ...      ...    ... 195   56    1   0            1   11.567  drugC 196   16    1   0            1   12.006  drugC 197   52    1   1            1    9.894  drugX 198   23    1   1            1   14.020  drugX 199   40    1   0            1   11.349  drugX  [200 rows x 6 columns]

We can observe that all "object" values ​​are processed into "binary" values ​​to represent categorical data. For example, in the cholesterol attribute, values ​​showing "low" are processed as 0, and "high" values ​​are processed as 1. Now we are ready to create the dependent and independent variables from the data.

Step 4: Split the Data

After processing our data into the correct structure, we now set the "X" variable (independent variable), "Y "Variable (dependent variable). Let us use python to implement

Python implementation:

X_var = df[['Sex', 'BP', 'Age', 'Cholesterol', 'Na_to_K']].values # 自变量 y_var = df['Drug'].values # 因变量  print(cl('X variable samples : {}'.format(X_var[:5]), attrs = ['bold'])) print(cl('Y variable samples : {}'.format(y_var[:5]), attrs = ['bold']))

Output:

X variable samples : [[ 1.     2.    23.     1.    25.355]  [ 1.     0.    47.     1.    13.093]  [ 1.     0.    47.     1.    10.114]  [ 1.     1.    28.     1.     7.798]  [ 1.     0.    61.     1.    18.043]] Y variable samples : ['drugY' 'drugC' 'drugC' 'drugX' 'drugY']

We can now use the "train_test_split" algorithm in scikit learn to The data is split into a training set and a test set, which contain the X and Y variables we defined. Follow the code to split data in python.

Python implementation:

X_train, X_test, y_train, y_test = train_test_split(X_var, y_var, test_size = 0.2, random_state = 0)  print(cl('X_train shape : {}'.format(X_train.shape), attrs = ['bold'], color = 'black')) print(cl('X_test shape : {}'.format(X_test.shape), attrs = ['bold'], color = 'black')) print(cl('y_train shape : {}'.format(y_train.shape), attrs = ['bold'], color = 'black')) print(cl('y_test shape : {}'.format(y_test.shape), attrs = ['bold'], color = 'black'))

Output:

X_train shape : (160, 5) X_test shape : (40, 5) y_train shape : (160,) y_test shape : (40,)

Now we have all the components to build the decision tree model. So, let's go ahead and build our model in python.

Step 5: Build model and prediction

With the help of the "DecisionTreeClassifier" algorithm provided by the scikit learning package, it is feasible to build a decision tree. Afterwards, we can use our trained model to predict our data. Finally, the accuracy of our prediction results can be calculated using the "Accuracy" evaluation metric. Let’s use python to complete this process!

Python implementation:

model = dtc(criterion = 'entropy', max_depth = 4) model.fit(X_train, y_train)  pred_model = model.predict(X_test)  print(cl('Accuracy of the model is {:.0%}'.format(accuracy_score(y_test, pred_model)), attrs = ['bold']))

Output:

Accuracy of the model is 88%

在代码的第一步中，我们定义了一个名为“model”变量的变量，我们在其中存储DecisionTreeClassifier模型。接下来，我们将使用我们的训练集对模型进行拟合和训练。之后，我们定义了一个变量，称为“pred_model”变量，其中我们将模型预测的所有值存储在数据上。最后，我们计算了我们的预测值与实际值的精度，其准确率为88%。

步骤6：可视化模型

现在我们有了决策树模型，让我们利用python中scikit  learn包提供的“plot_tree”函数来可视化它。按照代码从python中的决策树模型生成一个漂亮的树图。

Python实现：

feature_names = df.columns[:5] target_names = df['Drug'].unique().tolist()  plot_tree(model,            feature_names = feature_names,            class_names = target_names,            filled = True,            rounded = True)  plt.savefig('tree_visualization.png')

输出：

结论

有很多技术和其他算法用于优化决策树和避免过拟合，比如剪枝。虽然决策树通常是不稳定的，这意味着数据的微小变化会导致最优树结构的巨大变化，但其简单性使其成为广泛应用的有力候选。在神经网络流行之前，决策树是机器学习中最先进的算法。其他一些集成模型，比如随机森林模型，比普通决策树模型更强大。

决策树由于其简单性和可解释性而非常强大。决策树和随机森林在用户注册建模、信用评分、故障预测、医疗诊断等领域有着广泛的应用。我为本文提供了完整的代码。

完整代码：

import pandas as pd # 数据处理 import numpy as np # 使用数组 import matplotlib.pyplot as plt # 可视化 from matplotlib import rcParams # 图大小 from termcolor import colored as cl # 文本自定义  from sklearn.tree import DecisionTreeClassifier as dtc # 树算法 from sklearn.model_selection import train_test_split # 拆分数据 from sklearn.metrics import accuracy_score # 模型准确度 from sklearn.tree import plot_tree # 树图  rcParams['figure.figsize'] = (25, 20)  df = pd.read_csv('drug.csv') df.drop('Unnamed: 0', axis = 1, inplace = True)  print(cl(df.head(), attrs = ['bold']))  df.info()  for i in df.Sex.values:     if i  == 'M':         df.Sex.replace(i, 0, inplace = True)     else:         df.Sex.replace(i, 1, inplace = True)  for i in df.BP.values:     if i == 'LOW':         df.BP.replace(i, 0, inplace = True)     elif i == 'NORMAL':         df.BP.replace(i, 1, inplace = True)     elif i == 'HIGH':         df.BP.replace(i, 2, inplace = True)  for i in df.Cholesterol.values:     if i == 'LOW':         df.Cholesterol.replace(i, 0, inplace = True)     else:         df.Cholesterol.replace(i, 1, inplace = True)  print(cl(df, attrs = ['bold']))  X_var = df[['Sex', 'BP', 'Age', 'Cholesterol', 'Na_to_K']].values # 自变量 y_var = df['Drug'].values # 因变量  print(cl('X variable samples : {}'.format(X_var[:5]), attrs = ['bold'])) print(cl('Y variable samples : {}'.format(y_var[:5]), attrs = ['bold']))  X_train, X_test, y_train, y_test = train_test_split(X_var, y_var, test_size = 0.2, random_state = 0)  print(cl('X_train shape : {}'.format(X_train.shape), attrs = ['bold'], color = 'red')) print(cl('X_test shape : {}'.format(X_test.shape), attrs = ['bold'], color = 'red')) print(cl('y_train shape : {}'.format(y_train.shape), attrs = ['bold'], color = 'green')) print(cl('y_test shape : {}'.format(y_test.shape), attrs = ['bold'], color = 'green'))  model = dtc(criterion = 'entropy', max_depth = 4) model.fit(X_train, y_train)  pred_model = model.predict(X_test)  print(cl('Accuracy of the model is {:.0%}'.format(accuracy_score(y_test, pred_model)), attrs = ['bold']))  feature_names = df.columns[:5] target_names = df['Drug'].unique().tolist()  plot_tree(model,            feature_names = feature_names,            class_names = target_names,            filled = True,            rounded = True)  plt.savefig('tree_visualization.png')

The above is the detailed content of How to build a decision tree in Python. For more information, please follow other related articles on the PHP Chinese website!