138 lines
4.3 KiB
Markdown
138 lines
4.3 KiB
Markdown
---
|
||
title: db1的机器学习笔记
|
||
createTime: 2025/02/24 03:13:20
|
||
permalink: /learning-notes/ml/personal/db1/
|
||
icon: /avatar/db1.jpg
|
||
---
|
||
|
||
编写者: [@db1](/friends/persons/)
|
||
## 代码示例
|
||
|
||
**监督机器学习的代码示例及步骤(以scikit-learn框架为例)：**
|
||
|
||
### 1. 数据预处理
|
||
|
||
**数据集导入**
|
||
|
||
```python
|
||
df = pd.read_excel('NHANES.xlsx')
|
||
```
|
||
|
||
**数据集的划分** 以常见的8：2划分数据集为训练集和测试集
|
||
|
||
```python
|
||
y=df['GDM']#分类变量
|
||
X=df[['Age.Class', 'BMI.Class', 'Pregnancy.times','Parturition.times','UA','ALT','AST','GGT','PLT','Ca','NPAR']]
|
||
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=42)
|
||
```
|
||
|
||
**是否需要 缺失值的插补**
|
||
|
||
**是否需要 平衡类别**
|
||
|
||
在这一步考虑训练集的类别是否平衡，是否需要进行平衡类别。例如对不平衡的数据集进行SMOTE过采样。
|
||
|
||
```python
|
||
smote = SMOTE(random_state=42)
|
||
|
||
X_resampled, y_resampled = smote.fit_resample(X_train, y_train)
|
||
|
||
print("平衡后数据集类别分布:", Counter(y_resampled))
|
||
|
||
X_train=X_resampled
|
||
y_train=y_resampled
|
||
```
|
||
|
||
**数据的归一化**
|
||
|
||
### 2. 模型选择
|
||
|
||
这里以常见的几种模型为例，并以分类问题为例，尝试得到其训练集和测试集的accuracy，AUC，并以一个表格的形式展现出来。
|
||
|
||
```python
|
||
# Model creation
|
||
List_of_models = [LogisticRegression(random_state=42),
|
||
KNeighborsClassifier(),
|
||
GaussianNB(),
|
||
DecisionTreeClassifier(random_state=42),
|
||
SVC(probability=True),
|
||
RandomForestClassifier(random_state=42)]
|
||
List_of_models_for_graph = ["LogisticRegression", "KNeighborsClassifier", "GaussianNB", "DecisionTreeClassifier","SVM", "RandomForestClassifier"]
|
||
df_results = pd.DataFrame(index = ["train accuracy", "test accuracy","train auc","test auc"])
|
||
|
||
for i in range(len(List_of_models)):
|
||
model_class = List_of_models[i]
|
||
model_class.fit(X_train, y_train)
|
||
y_scores = model_class.predict_proba(X_train)[:, 1] # 获取正类的预测概率
|
||
auc1 = roc_auc_score(y_train, y_scores)
|
||
y_scores = model_class.predict_proba(X_test)[:, 1] # 获取正类的预测概率
|
||
auc2 = roc_auc_score(y_test, y_scores)
|
||
results_classification = np.array([model_class.score(X_train,y_train), model_class.score(X_test,y_test),auc1,auc2])
|
||
df_results[List_of_models_for_graph[i]] = results_classification
|
||
df_graph = df_results.transpose()#行列交换，转置
|
||
```
|
||
|
||
|
||
|
||
### 3. 调参
|
||
|
||
常用的调参方式为网格搜索，optuna。以下为对随机森林进行网格搜索调参的代码：
|
||
|
||
```python
|
||
RandomForestClassifier_parameters = {'criterion' : ['gini', 'entropy'],
|
||
'n_estimators' : [1,10,20,30, 100, 200, 400],
|
||
'min_samples_split' : [7,9,10,20,30],
|
||
'min_samples_leaf' : [1,2,5,7,9,10],
|
||
'max_features' : ['sqrt', 'log2'],
|
||
}
|
||
|
||
|
||
RandomForestClassifier_GridSearchCV = GridSearchCV(estimator = RandomForestClassifier(random_state=42), param_grid = RandomForestClassifier_parameters, cv=5, n_jobs=-1, verbose=0
|
||
)
|
||
RandomForestClassifier_GridSearchCV.fit(X_train, y_train)
|
||
List_of_models[5]=RandomForestClassifier_GridSearchCV.best_estimator_#更新Model列表
|
||
print(RandomForestClassifier_GridSearchCV.best_params_)
|
||
```
|
||
|
||
|
||
|
||
### 4. 结果显示
|
||
|
||
例如：以ROC作图的形式输出结果
|
||
|
||
```python
|
||
models = {
|
||
"KNeighbors Classifier": List_of_models[1],
|
||
"GaussianNB": List_of_models[2],
|
||
"DecisionTree Classifier": List_of_models[3],
|
||
"SVM": List_of_models[4],
|
||
"RF": List_of_models[5],
|
||
}
|
||
|
||
# 绘制ROC曲线
|
||
plt.figure(figsize=(10, 10))
|
||
|
||
for name, model in models.items():
|
||
y_scores = model.predict_proba(X_test)[:, 1] # 获取正类的概率
|
||
fpr, tpr, _ = roc_curve(y_test, y_scores)
|
||
roc_auc = auc(fpr, tpr)
|
||
plt.plot(fpr, tpr, label=f'{name} (AUC = {roc_auc:.3f})')
|
||
|
||
plt.plot([0, 1], [0, 1], 'k--', label='Random Guess')
|
||
|
||
plt.title('ROC Curve for Testing Set')
|
||
plt.xlabel('False Positive Rate')
|
||
plt.ylabel('True Positive Rate')
|
||
plt.legend(loc='lower right')
|
||
plt.xlim([0.0, 1.0])
|
||
plt.ylim([0.0, 1.0])
|
||
plt.grid()
|
||
|
||
plt.show()
|
||
```
|
||
|
||
------
|
||
|
||
### 5. 完整代码下载
|
||
|
||
<RepoCard repo="db1-bot/mL-learning-notes"></RepoCard> |