Revenue-Optimization-Based-on-Stackelberg-Game-Theory/model_building.py at main · OscarXie147/Revenue-Optimization-Based-on-Stackelberg-Game-Theory · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
import pandas as pd
import numpy as np
import lightgbm as lgb
from sklearn.metrics import roc_auc_score, mean_squared_error
import warnings

warnings.filterwarnings('ignore', category=UserWarning)

class DynamicPricingModeler:
    """
    负责构建、训练和评估动态定价所需的混合模型。
    """
    def __init__(self, processed_data: pd.DataFrame, large_order_threshold: int = 500):
        if not isinstance(processed_data, pd.DataFrame) or processed_data.empty:
            raise ValueError("需要提供一个非空的、经过预处理的Pandas DataFrame。")

        self.df = processed_data
        self.LARGE_ORDER_THRESHOLD = large_order_threshold
        print(f"“大额订单”阈值设定为: quantity >= {self.LARGE_ORDER_THRESHOLD}")

        g_success_count = self.df['is_success'].value_counts().get(1, 1)
        g_rejection_count = self.df['is_success'].value_counts().get(0, 1)
        g_scale_weight = g_success_count / g_rejection_count
        self.g_model = lgb.LGBMClassifier(random_state=42, objective='binary', metric='auc', scale_pos_weight=g_scale_weight)

        self.f_model_clf = lgb.LGBMClassifier(random_state=42, objective='binary', metric='auc')
        self.f_model_reg = lgb.LGBMRegressor(random_state=42, objective='regression_l1', metric='rmse')

        print("动态定价建模器已初始化，准备构建混合模型。")

    def build_follower_response_models(self):
        print("\n--- 开始构建阶段二：跟随者响应模型 ---")

        # 更新：使用新的、无数据泄露的特征列表
        features = [
            'price', 'stockcode', 'customer_id', 'month', 'weekday', 'hour',
            'customer_trade_count', 'customer_cancellation_rate',
            'product_trade_count', 'product_rejection_rate', 'price_deviation'
        ]
        categorical_features = ['stockcode', 'customer_id']

        # --- 模型一 (g_model): 交易成功率分类模型 ---
        print("\n[1/3] 正在构建 g_model (交易成功率分类模型)...")
        train_df = self.df[self.df['year'] < 2011]
        test_df = self.df[self.df['year'] == 2011]
        X_train_g, y_train_g = train_df[features], train_df['is_success']
        X_test_g, y_test_g = test_df[features], test_df['is_success']
        self.g_model.fit(X_train_g, y_train_g, categorical_feature=categorical_features,
                         eval_set=[(X_test_g, y_test_g)], callbacks=[lgb.early_stopping(10, verbose=False)])
        y_pred_proba_g = self.g_model.predict_proba(X_test_g)[:, 1]
        auc_score_g = roc_auc_score(y_test_g, y_pred_proba_g)
        print(f"g_model 在测试集上的评估完成。ROC-AUC Score: {auc_score_g:.4f}")

        # --- F-Model 混合策略 ---
        success_df = self.df[self.df['is_success'] == 1].copy()
        success_df['is_large_order'] = (success_df['quantity'] >= self.LARGE_ORDER_THRESHOLD).astype(int)

        train_success_df = success_df[success_df['year'] < 2011]
        test_success_df = success_df[success_df['year'] == 2011]

        # --- 模型二 (f_model_clf): 大额订单分类器 ---
        print("\n[2/3] 正在构建 f_model_clf (大额订单分类器)...")
        target_f_clf = 'is_large_order'
        X_train_f_clf, y_train_f_clf = train_success_df[features], train_success_df[target_f_clf]
        X_test_f_clf, y_test_f_clf = test_success_df[features], test_success_df[target_f_clf]

        clf_pos_count = y_train_f_clf.value_counts().get(0, 1)
        clf_neg_count = y_train_f_clf.value_counts().get(1, 1)
        clf_scale_weight = clf_pos_count / clf_neg_count
        self.f_model_clf.set_params(scale_pos_weight=clf_scale_weight)

        self.f_model_clf.fit(X_train_f_clf, y_train_f_clf, categorical_feature=categorical_features,
                             eval_set=[(X_test_f_clf, y_test_f_clf)], callbacks=[lgb.early_stopping(10, verbose=False)])

        y_pred_proba_f_clf = self.f_model_clf.predict_proba(X_test_f_clf)[:, 1]
        auc_score_f_clf = roc_auc_score(y_test_f_clf, y_pred_proba_f_clf)
        print(f"f_model_clf 在测试集上的评估完成。ROC-AUC Score: {auc_score_f_clf:.4f}")

        # --- 模型三 (f_model_reg): 常规订单回归器 ---
        print("\n[3/3] 正在构建 f_model_reg (常规订单回归器)...")

        regular_orders_train = train_success_df[train_success_df['is_large_order'] == 0].copy()
        regular_orders_test = test_success_df[test_success_df['is_large_order'] == 0].copy()

        regular_orders_train['quantity_log'] = np.log1p(regular_orders_train['quantity'])

        X_train_f_reg, y_train_f_reg = regular_orders_train[features], regular_orders_train['quantity_log']
        X_test_f_reg, y_test_f_reg_actual = regular_orders_test[features], regular_orders_test['quantity']

        self.f_model_reg.fit(X_train_f_reg, y_train_f_reg, categorical_feature=categorical_features,
                             eval_set=[(X_test_f_reg, np.log1p(y_test_f_reg_actual))],
                             callbacks=[lgb.early_stopping(10, verbose=False)])

        y_pred_log_f_reg = self.f_model_reg.predict(X_test_f_reg)
        y_pred_f_reg = np.expm1(y_pred_log_f_reg)
        y_pred_f_reg[y_pred_f_reg < 0] = 0

        rmse_reg = np.sqrt(mean_squared_error(y_test_f_reg_actual, y_pred_f_reg))
        print(f"f_model_reg 在【常规订单】测试集上的评估完成。RMSE: {rmse_reg:.4f}")

        print("\n--- 阶段二模型构建全部完成 ---")
        return self.g_model, self.f_model_clf, self.f_model_reg
    def train_final_models(self):
        """
        在全部数据集上训练最终的模型，不进行划分，用于最终的优化。
        """
        print("在全部数据上训练最终模型...")

        features = [
            'price', 'stockcode', 'customer_id', 'month', 'weekday', 'hour',
            'customer_trade_count', 'customer_cancellation_rate',
            'product_trade_count', 'product_rejection_rate', 'price_deviation'
        ]
        categorical_features = ['stockcode', 'customer_id']

        # 训练最终的 g_model
        X_g, y_g = self.df[features], self.df['is_success']
        self.g_model.fit(X_g, y_g, categorical_feature=categorical_features)
        print("最终 g_model 训练完成。")

        # 准备f-model的数据
        success_df = self.df[self.df['is_success'] == 1].copy()
        success_df['is_large_order'] = (success_df['quantity'] >= self.LARGE_ORDER_THRESHOLD).astype(int)

        # 训练最终的 f_model_clf
        X_f_clf, y_f_clf = success_df[features], success_df['is_large_order']
        self.f_model_clf.fit(X_f_clf, y_f_clf, categorical_feature=categorical_features)
        print("最终 f_model_clf 训练完成。")

        # 训练最终的 f_model_reg
        regular_orders = success_df[success_df['is_large_order'] == 0].copy()
        regular_orders['quantity_log'] = np.log1p(regular_orders['quantity'])
        X_f_reg, y_f_reg = regular_orders[features], regular_orders['quantity_log']
        self.f_model_reg.fit(X_f_reg, y_f_reg, categorical_feature=categorical_features)
        print("最终 f_model_reg 训练完成。")

        return self.g_model, self.f_model_clf, self.f_model_reg