AutoML 自动机器学习¶

原始数据 → [自动预处理] → [自动特征工程] → [自动模型选择] → [自动超参调优] → 最优模型

import numpy as np
import pandas as pd
from sklearn.datasets import make_classification
from sklearn.model_selection import train_test_split, cross_val_score
from sklearn.metrics import roc_auc_score

# ===== 1. 用 Optuna 做超参优化（最灵活的 AutoML 工具）=====
# pip install optuna
import optuna
optuna.logging.set_verbosity(optuna.logging.WARNING)  # 减少日志输出

from sklearn.ensemble import RandomForestClassifier, GradientBoostingClassifier
from sklearn.svm import SVC

# 模拟癌症分类数据（基因表达 + 标签）
X, y = make_classification(
    n_samples=300,
    n_features=100,    # 100 个基因特征
    n_informative=20,  # 20 个真正重要的基因
    n_redundant=10,
    random_state=42
)

X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=42)

def objective(trial):
    """
    Optuna 目标函数：定义搜索空间，返回要优化的指标
    trial 对象自动管理超参采样和历史记录
    """
    # 选择模型类型（离散选择）
    classifier_name = trial.suggest_categorical(
        'classifier', ['RandomForest', 'GradientBoosting', 'SVM']
    )

    if classifier_name == 'RandomForest':
        # 随机森林超参
        n_estimators = trial.suggest_int('n_estimators', 50, 300)   # 树的数量
        max_depth = trial.suggest_int('max_depth', 3, 15)            # 最大深度
        min_samples_split = trial.suggest_int('min_samples_split', 2, 10)
        clf = RandomForestClassifier(
            n_estimators=n_estimators,
            max_depth=max_depth,
            min_samples_split=min_samples_split,
            random_state=42
        )

    elif classifier_name == 'GradientBoosting':
        # 梯度提升超参
        n_estimators = trial.suggest_int('gb_n_estimators', 50, 200)
        learning_rate = trial.suggest_float('learning_rate', 0.01, 0.3, log=True)  # 对数采样
        max_depth = trial.suggest_int('gb_max_depth', 2, 8)
        clf = GradientBoostingClassifier(
            n_estimators=n_estimators,
            learning_rate=learning_rate,
            max_depth=max_depth,
            random_state=42
        )

    else:  # SVM
        C = trial.suggest_float('C', 0.1, 100, log=True)       # 正则化参数
        kernel = trial.suggest_categorical('kernel', ['rbf', 'linear'])
        clf = SVC(C=C, kernel=kernel, probability=True, random_state=42)

    # 用 5 折交叉验证评估（防止过拟合）
    scores = cross_val_score(clf, X_train, y_train, cv=5, scoring='roc_auc')
    return scores.mean()  # 返回平均 AUC（Optuna 默认最大化）

# 创建 Optuna 学习计划（study）
study = optuna.create_study(direction='maximize')  # 最大化 AUC
study.optimize(objective, n_trials=50)             # 尝试 50 种超参组合

print(f'最佳 AUC: {study.best_value:.4f}')
print(f'最佳超参: {study.best_params}')

# 用最佳超参重新训练并评估
best_params = study.best_params
classifier_name = best_params.pop('classifier')

if classifier_name == 'RandomForest':
    # 移除其他模型的参数
    rf_params = {k: v for k, v in best_params.items()
                 if k in ['n_estimators', 'max_depth', 'min_samples_split']}
    best_model = RandomForestClassifier(**rf_params, random_state=42)
else:
    best_model = RandomForestClassifier(random_state=42)  # 简化处理

best_model.fit(X_train, y_train)
y_pred_proba = best_model.predict_proba(X_test)[:, 1]
test_auc = roc_auc_score(y_test, y_pred_proba)
print(f'测试集 AUC: {test_auc:.4f}')

# ===== 2. PyCaret 低代码 AutoML 示例 =====
# pip install pycaret
try:
    from pycaret.classification import setup, compare_models, pull, finalize_model

    # PyCaret 的工作流（只需几行代码）
    df = pd.DataFrame(X_train, columns=[f'gene_{i}' for i in range(100)])
    df['label'] = y_train

    # 初始化实验（自动预处理、交叉验证设置）
    exp = setup(
        data=df,
        target='label',         # 目标列名
        session_id=42,           # 随机种子
        verbose=False
    )

    # 比较所有可用模型，选出最佳
    best_models = compare_models(n_select=3, sort='AUC', verbose=False)
    results = pull()  # 获取比较结果表格
    print('\nPyCaret 模型比较结果（前5）:')
    print(results[['Model', 'AUC', 'F1']].head())

except ImportError:
    print("PyCaret 未安装。运行: pip install pycaret")

# ===== 3. 贝叶斯优化可视化 =====
import optuna.visualization as vis
try:
    # 优化历史图
    fig = vis.plot_optimization_history(study)
    print('\n贝叶斯优化收敛情况：可通过 fig.show() 查看')

    # 特征重要性（超参对结果的影响）
    fig2 = vis.plot_param_importances(study)
    print('超参重要性：可通过 fig2.show() 查看')
except Exception as e:
    print(f'可视化需要 plotly: {e}')