1. 数据集介绍¶

数据集共包含 11 个特征，共计 3269 条数据. 我们通过训练模型来预测红酒的品质, 品质共有 6 个各类别，分别使用数字: 0、1、2、3、4、5 来表示。

2. 案例实现¶

3.1 导入需要的库文件¶

import itertools
import joblib
import numpy as np
import xgboost as xgb
import pandas as pd
import numpy as np
from collections import Counter
from imblearn.over_sampling import SMOTE
from sklearn.model_selection import train_test_split
from sklearn.metrics import classification_report
from sklearn.model_selection import StratifiedKFold
from sklearn.base import clone
import time


pd.set_option('display.max_columns', None)
pd.set_option('display.max_rows', None)
pd.set_option('display.width', 1000)

3.2 数据基本处理¶

def test01():

    # 1. 加载训练数据
    data = pd.read_csv('data/红酒品质分类.csv')
    x = data.iloc[:, :-1]
    y = data.iloc[:, -1] - 3

    # 2. 不平衡数据处理
    print('数据类别比例:', Counter(y))
    x, y = SMOTE().fit_resample(x, y)
    print('数据类别比例:', Counter(y))

    # 3. 数据集分割
    x_train, x_valid, y_train, y_valid = train_test_split(x, y, test_size=0.2, stratify=y, random_state=22)

    # 4. 存储数据
    pd.concat([x_train, y_train], axis=1).to_csv('data/红酒品质分类-train.csv')
    pd.concat([x_valid, y_valid], axis=1).to_csv('data/红酒品质分类-valid.csv')

3.3 模型基本训练¶

def test02():

    # 1. 加载训练数据
    train_data = pd.read_csv('data/红酒品质分类-train.csv')
    valid_data = pd.read_csv('data/红酒品质分类-valid.csv')

    # 训练集
    x_train = train_data.iloc[:, :-1]
    y_train = train_data.iloc[:, -1]

    # 测试集
    x_valid = valid_data.iloc[:, :-1]
    y_valid = valid_data.iloc[:, -1]

    # 2. XGBoost模型训练
    estimator = xgb.XGBClassifier(n_estimators=100,
                                  objective='multi:softmax',
                                  eval_metric='merror',
                                  eta=0.1,
                                  use_label_encoder=False,
                                  random_state=22)
    estimator.fit(x_train, y_train)

    # 3. 模型评估
    y_pred = estimator.predict(x_valid)
    print(classification_report(y_true=y_valid, y_pred=y_pred))

    # 4. 模型保存
    joblib.dump(estimator, 'model/xgboost.pth')

3.4 模型参数调优¶

def test03():

    # 1. 加载训练数据
    train_data = pd.read_csv('data/红酒品质分类-train.csv')
    valid_data = pd.read_csv('data/红酒品质分类-valid.csv')

    # 训练集
    x_train = train_data.iloc[:, :-1]
    y_train = train_data.iloc[:, -1]

    # 测试集
    x_valid = valid_data.iloc[:, :-1]
    y_valid = valid_data.iloc[:, -1]


    # 2. 定义超参数
    param_grid = {'max_depth': np.arange(3, 11, 1),
                  'n_estimators': np.arange(50, 550, 50),
                  'eta': np.arange(0.1, 1, 0.1)}

    # 3. 超参数组合
    params = param_grid.items()
    params_start = [0] * len(params)
    params_end = [len(value) - 1 for _, value in params]
    param_list = []
    while True:

        input_params = {}
        for index, (key, value) in zip(params_start, params):
            input_params[key] = value[index]
        param_list.append(input_params)

        if params_start == params_end:
            break

        params_start[0] += 1
        for idx in range(len(params_start) - 1):
            if params_start[idx] > params_end[idx]:
                params_start[idx] = 0
                params_start[idx + 1] += 1


    # 4. 交叉验证分割数据集
    train_data = []
    spliter = StratifiedKFold(n_splits=5, shuffle=True)
    for train_index, valid_index in spliter.split(x_train, y_train):
        train_data.append({'train': train_index, 'valid': valid_index})

    # 5. 寻找找参数组合
    train_result = []
    for param in param_list:

        start = time.time()

        # 初始化模型
        base_estimator = xgb.XGBClassifier(use_label_encoder=False, eval_metric='merror')
        # 存储模型评分
        train_score = []
        # 模型训练和评估
        for data in train_data:
            # 获得训练和验证集
            train_index, valid_index = data['train'], data['valid']
            train_x, train_y = x_train.values[train_index], y_train.values[train_index]
            valid_x, valid_y = x_train.values[valid_index], y_train.values[valid_index]

            # 模型训练
            estimator = clone(base_estimator)
            estimator.set_params(**param)
            estimator.fit(train_x, train_y)

            # 模型评估
            score = estimator.score(valid_x, valid_y)
            train_score.append(score)

        # 打印模型平均分
        train_result.append([np.mean(train_score), param])

        end = time.time()
        print('score: %.2f paramters: %s time: %.2fs' % (np.mean(train_score), param, end - start))

    # 训练结果排序
    train_result.sort(key=lambda x: x[0], reverse=True)

    # 选择最优参数训练最终模型
    best_param = train_result[0][1]
    print('best_param:', best_param)
    best_estimator = clone(base_estimator)
    best_estimator.set_params(**best_param)
    best_estimator.fit(x_train, y_train)

    # 存储最优模型
    joblib.dump(best_estimator, 'model/best_xgboost.pth')

    # 使用最优模型进行预测
    acc = best_estimator.score(x_valid, y_valid)
    print('Accurary:', acc)