ins-pricing 0.1.6__py3-none-any.whl

user_packages legacy/Try/BayesOptV7.py

@@ -0,0 +1,2832 @@
+# 数据在CPU和GPU之间传输会带来较大开销,但可以多CUDA流同时传输数据和计算,从而实现更大数据集的操作。
+
+import copy
+import gc
+import math
+import os
+from dataclasses import dataclass
+from pathlib import Path
+from re import X
+from typing import Any, Dict, List, Optional
+import csv
+
+import joblib
+import matplotlib.pyplot as plt
+import numpy as np  # 1.26.2
+import optuna  # 4.3.0
+import pandas as pd  # 2.2.3
+import shap
+import statsmodels.api as sm
+
+import torch  # 版本: 1.10.1+cu111
+import torch.nn as nn
+import torch.nn.functional as F
+import xgboost as xgb  # 1.7.0
+
+from torch.utils.data import Dataset, DataLoader, TensorDataset
+from torch.cuda.amp import autocast, GradScaler
+from torch.nn.utils import clip_grad_norm_
+from sklearn.model_selection import ShuffleSplit, cross_val_score  # 1.2.2
+from sklearn.preprocessing import StandardScaler
+from sklearn.metrics import log_loss, make_scorer, mean_tweedie_deviance
+
+# =============================================================================
+# Constants & utilities
+# =============================================================================
+EPS = 1e-8
+
+def csv_to_dict(file_path):
+   with open(file_path, mode='r', encoding='utf-8') as file:
+       reader = csv.DictReader(file)
+       data = [dict(filter(lambda item: item[0] != '', row.items())) for row in reader]
+   return data
+
+def ensure_parent_dir(file_path: str) -> None:
+    # 若目标文件所在目录不存在则自动创建
+    directory = os.path.dirname(file_path)
+    if directory:
+        os.makedirs(directory, exist_ok=True)
+
+
+def compute_batch_size(data_size: int, learning_rate: float, batch_num: int, minimum: int) -> int:
+    # 按学习率和样本量给出估算 batch，再夹在 [1, data_size] 范围内
+    estimated = int((learning_rate / 1e-4) ** 0.5 *
+                    (data_size / max(batch_num, 1)))
+    return max(1, min(data_size, max(minimum, estimated)))
+
+
+# 定义在 PyTorch 环境下的 Tweedie 偏差损失函数
+# 参考文档：https://scikit-learn.org/stable/modules/model_evaluation.html#mean-poisson-gamma-and-tweedie-deviances
+
+
+def tweedie_loss(pred, target, p=1.5, eps=1e-6, max_clip=1e6):
+    # 为确保稳定性先将预测值裁剪为正数
+    pred_clamped = torch.clamp(pred, min=eps)
+    # 计算 Tweedie 偏差的各部分
+    if p == 1:
+        # 对应泊松分布
+        term1 = target * torch.log(target / pred_clamped + eps)
+        term2 = -target + pred_clamped
+        term3 = 0
+    elif p == 0:
+        # 对应高斯分布
+        term1 = 0.5 * torch.pow(target - pred_clamped, 2)
+        term2 = 0
+        term3 = 0
+    elif p == 2:
+        # 对应伽马分布
+        term1 = torch.log(pred_clamped / target + eps)
+        term2 = -target / pred_clamped + 1
+        term3 = 0
+    else:
+        term1 = torch.pow(target, 2 - p) / ((1 - p) * (2 - p))
+        term2 = target * torch.pow(pred_clamped, 1 - p) / (1 - p)
+        term3 = torch.pow(pred_clamped, 2 - p) / (2 - p)
+    # Tweedie 负对数似然（忽略常数项）
+    return torch.nan_to_num(2 * (term1 - term2 + term3), nan=eps, posinf=max_clip, neginf=-max_clip)
+
+# 定义释放CUDA内存函数
+
+
+def free_cuda():
+    print(">>> Moving all models to CPU...")
+    for obj in gc.get_objects():
+        try:
+            if hasattr(obj, "to") and callable(obj.to):
+                # 跳过 torch.device 等不可移动对象
+                obj.to("cpu")
+        except:
+            pass
+
+    print(">>> Deleting tensors, optimizers, dataloaders...")
+    gc.collect()
+
+    print(">>> Emptying CUDA cache...")
+    torch.cuda.empty_cache()
+    torch.cuda.synchronize()
+
+    print(">>> CUDA memory freed.")
+
+
+# =============================================================================
+# Plotting helpers
+# =============================================================================
+
+# 定义分箱函数
+
+
+def split_data(data, col_nme, wgt_nme, n_bins=10):
+    # 避免修改原始数据帧，先创建排序后的副本
+    data_sorted = data.sort_values(by=col_nme, ascending=True).copy()
+    data_sorted['cum_weight'] = data_sorted[wgt_nme].cumsum()
+    w_sum = data_sorted[wgt_nme].sum()
+    if w_sum <= EPS:
+        data_sorted.loc[:, 'bins'] = 0
+    else:
+        data_sorted.loc[:, 'bins'] = np.floor(
+            data_sorted['cum_weight'] * float(n_bins) / w_sum
+        )
+        data_sorted.loc[(data_sorted['bins'] == n_bins), 'bins'] = n_bins - 1
+    return data_sorted.groupby(['bins'], observed=True).sum(numeric_only=True)
+
+# 定义提纯曲线（Lift）绘制函数
+
+
+def plot_lift_list(pred_model, w_pred_list, w_act_list,
+                   weight_list, tgt_nme, n_bins=10,
+                   fig_nme='Lift Chart'):
+    lift_data = pd.DataFrame()
+    lift_data.loc[:, 'pred'] = pred_model
+    lift_data.loc[:, 'w_pred'] = w_pred_list
+    lift_data.loc[:, 'act'] = w_act_list
+    lift_data.loc[:, 'weight'] = weight_list
+    plot_data = split_data(lift_data, 'pred', 'weight', n_bins)
+    plot_data['exp_v'] = plot_data['w_pred'] / plot_data['weight']
+    plot_data['act_v'] = plot_data['act'] / plot_data['weight']
+    plot_data.reset_index(inplace=True)
+    fig = plt.figure(figsize=(7, 5))
+    ax = fig.add_subplot(111)
+    ax.plot(plot_data.index, plot_data['act_v'],
+            label='Actual', color='red')
+    ax.plot(plot_data.index, plot_data['exp_v'],
+            label='Predicted', color='blue')
+    ax.set_title(
+        'Lift Chart of %s' % tgt_nme, fontsize=8)
+    plt.xticks(plot_data.index,
+               plot_data.index,
+               rotation=90, fontsize=6)
+    plt.yticks(fontsize=6)
+    plt.legend(loc='upper left',
+               fontsize=5, frameon=False)
+    plt.margins(0.05)
+    ax2 = ax.twinx()
+    ax2.bar(plot_data.index, plot_data['weight'],
+            alpha=0.5, color='seagreen',
+            label='Earned Exposure')
+    plt.yticks(fontsize=6)
+    plt.legend(loc='upper right',
+               fontsize=5, frameon=False)
+    plt.subplots_adjust(wspace=0.3)
+    save_path = os.path.join(
+        os.getcwd(), 'plot', f'05_{tgt_nme}_{fig_nme}.png')
+    ensure_parent_dir(save_path)
+    plt.savefig(save_path, dpi=300)
+    plt.close(fig)
+
+# 定义双提纯曲线绘制函数
+
+
+def plot_dlift_list(pred_model_1, pred_model_2,
+                    model_nme_1, model_nme_2,
+                    tgt_nme,
+                    w_list, w_act_list, n_bins=10,
+                    fig_nme='Double Lift Chart'):
+    lift_data = pd.DataFrame()
+    lift_data.loc[:, 'pred1'] = pred_model_1
+    lift_data.loc[:, 'pred2'] = pred_model_2
+    lift_data.loc[:, 'diff_ly'] = lift_data['pred1'] / lift_data['pred2']
+    lift_data.loc[:, 'act'] = w_act_list
+    lift_data.loc[:, 'weight'] = w_list
+    lift_data.loc[:, 'w_pred1'] = lift_data['pred1'] * lift_data['weight']
+    lift_data.loc[:, 'w_pred2'] = lift_data['pred2'] * lift_data['weight']
+    plot_data = split_data(lift_data, 'diff_ly', 'weight', n_bins)
+    plot_data['exp_v1'] = plot_data['w_pred1'] / plot_data['act']
+    plot_data['exp_v2'] = plot_data['w_pred2'] / plot_data['act']
+    plot_data['act_v'] = plot_data['act']/plot_data['act']
+    plot_data.reset_index(inplace=True)
+    fig = plt.figure(figsize=(7, 5))
+    ax = fig.add_subplot(111)
+    ax.plot(plot_data.index, plot_data['act_v'],
+            label='Actual', color='red')
+    ax.plot(plot_data.index, plot_data['exp_v1'],
+            label=model_nme_1, color='blue')
+    ax.plot(plot_data.index, plot_data['exp_v2'],
+            label=model_nme_2, color='black')
+    ax.set_title(
+        'Double Lift Chart of %s' % tgt_nme, fontsize=8)
+    plt.xticks(plot_data.index,
+               plot_data.index,
+               rotation=90, fontsize=6)
+    plt.xlabel('%s / %s' % (model_nme_1, model_nme_2), fontsize=6)
+    plt.yticks(fontsize=6)
+    plt.legend(loc='upper left',
+               fontsize=5, frameon=False)
+    plt.margins(0.1)
+    plt.subplots_adjust(bottom=0.25, top=0.95, right=0.8)
+    ax2 = ax.twinx()
+    ax2.bar(plot_data.index, plot_data['weight'],
+            alpha=0.5, color='seagreen',
+            label='Earned Exposure')
+    plt.yticks(fontsize=6)
+    plt.legend(loc='upper right',
+               fontsize=5, frameon=False)
+    plt.subplots_adjust(wspace=0.3)
+    save_path = os.path.join(
+        os.getcwd(), 'plot', f'06_{tgt_nme}_{fig_nme}.png')
+    ensure_parent_dir(save_path)
+    plt.savefig(save_path, dpi=300)
+    plt.close(fig)
+
+
+# =============================================================================
+# ResNet model & sklearn-style wrapper
+# =============================================================================
+
+# 开始定义ResNet模型结构
+# 残差块：两层线性 + ReLU + 残差连接
+# ResBlock 继承 nn.Module
+class ResBlock(nn.Module):
+    def __init__(self, dim: int, dropout: float = 0.1,
+                 use_layernorm: bool = False, residual_scale: float = 0.1
+                 ):
+        super().__init__()
+        self.use_layernorm = use_layernorm
+
+        if use_layernorm:
+            Norm = nn.LayerNorm      # 对最后一维做归一化
+        else:
+            def Norm(d): return nn.BatchNorm1d(d)  # 保留一个开关,想试 BN 时也能用
+
+        self.norm1 = Norm(dim)
+        self.fc1 = nn.Linear(dim, dim, bias=True)
+        self.act = nn.ReLU(inplace=True)
+        self.dropout = nn.Dropout(dropout) if dropout > 0.0 else nn.Identity()
+        self.norm2 = Norm(dim)
+        self.fc2 = nn.Linear(dim, dim, bias=True)
+
+        # 残差缩放,防止一开始就把主干搞炸
+        self.res_scale = nn.Parameter(
+            torch.tensor(residual_scale, dtype=torch.float32)
+        )
+
+    def forward(self, x):
+        # 前置激活结构
+        out = self.norm1(x)
+        out = self.fc1(out)
+        out = self.act(out)
+        out = self.dropout(out)
+        out = self.norm2(out)
+        out = self.fc2(out)
+        # 残差缩放再相加
+        return F.relu(x + self.res_scale * out)
+
+# ResNetSequential 继承 nn.Module,定义整个网络结构
+
+
+class ResNetSequential(nn.Module):
+    # 输入张量形状：(batch, input_dim)
+    # 网络结构：全连接 + 归一化 + ReLU，再堆叠若干残差块，最后输出 Softplus
+
+    def __init__(self, input_dim: int, hidden_dim: int = 64, block_num: int = 2,
+                 use_layernorm: bool = True, dropout: float = 0.1,
+                 residual_scale: float = 0.1):
+        super(ResNetSequential, self).__init__()
+
+        self.net = nn.Sequential()
+        self.net.add_module('fc1', nn.Linear(input_dim, hidden_dim))
+
+        if use_layernorm:
+            self.net.add_module('norm1', nn.LayerNorm(hidden_dim))
+        else:
+            self.net.add_module('norm1', nn.BatchNorm1d(hidden_dim))
+
+        self.net.add_module('relu1', nn.ReLU(inplace=True))
+
+        # 多个残差块
+        for i in range(block_num):
+            self.net.add_module(
+                f'ResBlk_{i+1}',
+                ResBlock(
+                    hidden_dim,
+                    dropout=dropout,
+                    use_layernorm=use_layernorm,
+                    residual_scale=residual_scale)
+            )
+
+        self.net.add_module('fc_out', nn.Linear(hidden_dim, 1))
+        self.net.add_module('softplus', nn.Softplus())
+
+    def forward(self, x):
+        return self.net(x)
+
+# 定义ResNet模型的Scikit-Learn接口类
+
+
+class ResNetSklearn(nn.Module):
+    def __init__(self, model_nme: str, input_dim: int, hidden_dim: int = 64,
+                 block_num: int = 2, batch_num: int = 100, epochs: int = 100,
+                 task_type: str = 'regression',
+                 tweedie_power: float = 1.5, learning_rate: float = 0.01, patience: int = 10,
+                 use_layernorm: bool = True, dropout: float = 0.1,
+                 residual_scale: float = 0.1,
+                 use_data_parallel: bool = True):
+        super(ResNetSklearn, self).__init__()
+
+        self.input_dim = input_dim
+        self.hidden_dim = hidden_dim
+        self.block_num = block_num
+        self.batch_num = batch_num
+        self.epochs = epochs
+        self.task_type = task_type
+        self.model_nme = model_nme
+        self.learning_rate = learning_rate
+        self.patience = patience
+        self.use_layernorm = use_layernorm
+        self.dropout = dropout
+        self.residual_scale = residual_scale
+
+        # 设备选择：cuda > mps > cpu
+        if torch.cuda.is_available():
+            self.device = torch.device('cuda')
+        elif torch.backends.mps.is_available():
+            self.device = torch.device('mps')
+        else:
+            self.device = torch.device('cpu')
+
+        # Tweedie 幂指数设定
+        if 'f' in self.model_nme:
+            self.tw_power = 1
+        elif 's' in self.model_nme:
+            self.tw_power = 2
+        else:
+            self.tw_power = tweedie_power
+
+        # 搭建网络（先在 CPU 上建好）
+        core = ResNetSequential(
+            self.input_dim,
+            self.hidden_dim,
+            self.block_num,
+            use_layernorm=self.use_layernorm,
+            dropout=self.dropout,
+            residual_scale=self.residual_scale
+        )
+
+        # 如果是分类任务，替换掉最后的 Softplus
+        if self.task_type == 'classification':
+            core.net.softplus = nn.Identity()
+
+        # ===== ⭐ 多卡支持：DataParallel =====
+        if use_data_parallel and (self.device.type == 'cuda') and (torch.cuda.device_count() > 1):
+            core = nn.DataParallel(core, device_ids=list(range(torch.cuda.device_count())))
+            # DataParallel 会把输入 scatter 到多卡上，但“主设备”仍然是 cuda:0
+            self.device = torch.device('cuda')
+
+        self.resnet = core.to(self.device)
+
+    def forward(self, x):
+        # 处理 SHAP 的 NumPy 输入
+        if isinstance(x, np.ndarray):
+            x_tensor = torch.tensor(x, dtype=torch.float32)
+        else:
+            x_tensor = x
+
+        x_tensor = x_tensor.to(self.device)
+        y_pred = self.resnet(x_tensor)
+        return y_pred
+
+    # ---------------- 训练 ----------------
+
+    def fit(self, X_train, y_train, w_train=None,
+            X_val=None, y_val=None, w_val=None):
+
+        # === 1. 训练集：先留在 CPU，交给 DataLoader 批量搬运到 GPU ===
+        X_tensor = torch.tensor(X_train.values, dtype=torch.float32)
+        y_tensor = torch.tensor(
+            y_train.values, dtype=torch.float32).view(-1, 1)
+        if w_train is not None:
+            w_tensor = torch.tensor(
+                w_train.values, dtype=torch.float32).view(-1, 1)
+        else:
+            w_tensor = torch.ones_like(y_tensor)
+
+        # === 2. 验证集：先在 CPU 上构造，后续一次性搬到目标设备 ===
+        has_val = X_val is not None and y_val is not None
+        if has_val:
+            X_val_tensor = torch.tensor(X_val.values, dtype=torch.float32)
+            y_val_tensor = torch.tensor(
+                y_val.values, dtype=torch.float32).view(-1, 1)
+            if w_val is not None:
+                w_val_tensor = torch.tensor(
+                    w_val.values, dtype=torch.float32).view(-1, 1)
+            else:
+                w_val_tensor = torch.ones_like(y_val_tensor)
+        else:
+            X_val_tensor = y_val_tensor = w_val_tensor = None
+
+        # === 3. 构建 DataLoader ===
+        dataset = TensorDataset(X_tensor, y_tensor, w_tensor)
+        batch_size = compute_batch_size(
+            data_size=len(dataset),
+            learning_rate=self.learning_rate,
+            batch_num=self.batch_num,
+            minimum=64
+        )
+        N = X_tensor.shape[0]
+
+        if self.device.type == 'cuda':
+            if N > 200_000:
+                base_bs = 4096
+            elif N > 50_000:
+                base_bs = 2048
+            else:
+                base_bs = 1024
+        else:
+            if N > 50_000:
+                base_bs = 1024
+            else:
+                base_bs = 512
+
+        # 物理 batch：显存安全为主
+        batch_size = min(batch_size, base_bs, N)
+
+        # ===== ⭐ 梯度累计设置 =====
+        # 目标“有效 batch 大小”：影响收敛稳定性
+        if self.device.type == 'cuda':
+            target_effective_bs = 8192   # 可以按需要改成 4096 / 16384
+        else:
+            target_effective_bs = 4096
+
+        accum_steps = max(1, target_effective_bs // batch_size)
+
+        dataloader = DataLoader(
+            dataset,
+            batch_size=batch_size,
+            shuffle=True,
+            num_workers=1,  # 表格数据通常 0~1 个线程即可
+            pin_memory=(self.device.type == 'cuda')
+        )
+
+        # === 4. 优化器与 AMP ===
+        self.optimizer = torch.optim.Adam(
+            self.resnet.parameters(), lr=self.learning_rate)
+        self.scaler = GradScaler(enabled=(self.device.type == 'cuda'))
+
+        # === 5. 早停机制 ===
+        best_loss, patience_counter = float('inf'), 0
+        best_model_state = None
+
+        # 若存在验证集则一次性搬到目标设备
+        if has_val:
+            X_val_dev = X_val_tensor.to(self.device, non_blocking=True)
+            y_val_dev = y_val_tensor.to(self.device, non_blocking=True)
+            w_val_dev = w_val_tensor.to(self.device, non_blocking=True)
+
+        # === 6. 训练循环 ===
+        for epoch in range(1, self.epochs + 1):
+            self.resnet.train()
+            self.optimizer.zero_grad()  # ⭐ 梯度在 epoch 内累积
+
+            for step, (X_batch, y_batch, w_batch) in enumerate(dataloader):
+
+                X_batch = X_batch.to(self.device, non_blocking=True)
+                y_batch = y_batch.to(self.device, non_blocking=True)
+                w_batch = w_batch.to(self.device, non_blocking=True)
+
+                with autocast(enabled=(self.device.type == 'cuda')):
+                    y_pred = self.resnet(X_batch)
+
+                    if self.task_type == 'classification':
+                        loss_fn = nn.BCEWithLogitsLoss(reduction='none')
+                        losses = loss_fn(y_pred, y_batch).view(-1)
+                    else:
+                        y_pred = torch.clamp(y_pred, min=1e-6)
+                        losses = tweedie_loss(
+                            y_pred, y_batch, p=self.tw_power
+                        ).view(-1)
+
+                    weighted_loss = (
+                        losses * w_batch.view(-1)
+                    ).sum() / torch.clamp(w_batch.sum(), min=EPS)
+
+                    # ⭐ 梯度累计：loss 除以 accum_steps
+                    loss_for_backward = weighted_loss / accum_steps
+
+                self.scaler.scale(loss_for_backward).backward()
+
+                # ⭐ 到了累积次数或者最后一个 batch，才真正 step 一次
+                if ((step + 1) % accum_steps == 0) or ((step + 1) == len(dataloader)):
+                    if self.device.type == 'cuda':
+                        self.scaler.unscale_(self.optimizer)
+                        clip_grad_norm_(self.resnet.parameters(), max_norm=1.0)
+
+                    self.scaler.step(self.optimizer)
+                    self.scaler.update()
+                    self.optimizer.zero_grad()
+
+            # === 7. 验证损失与早停判断 ===
+            if has_val:
+                self.resnet.eval()
+                with torch.no_grad(), autocast(enabled=(self.device.type == 'cuda')):
+                    y_val_pred = self.resnet(X_val_dev)
+
+                    if self.task_type == 'classification':
+                        val_loss_fn = nn.BCEWithLogitsLoss(reduction='none')
+                        val_loss_values = val_loss_fn(
+                            y_val_pred, y_val_dev).view(-1)
+                    else:
+                        y_val_pred = torch.clamp(y_val_pred, min=1e-6)
+                        val_loss_values = tweedie_loss(
+                            y_val_pred, y_val_dev, p=self.tw_power
+                        ).view(-1)
+
+                    val_weighted_loss = (
+                        val_loss_values * w_val_dev.view(-1)
+                    ).sum() / torch.clamp(w_val_dev.sum(), min=EPS)
+
+                if val_weighted_loss < best_loss:
+                    best_loss = val_weighted_loss
+                    patience_counter = 0
+                    best_model_state = copy.deepcopy(self.resnet.state_dict())
+                else:
+                    patience_counter += 1
+
+                if patience_counter >= self.patience and best_model_state is not None:
+                    self.resnet.load_state_dict(best_model_state)
+                    break
+
+        if has_val and best_model_state is not None:
+            self.resnet.load_state_dict(best_model_state)
+
+    # ---------------- 预测 ----------------
+
+    def predict(self, X_test):
+        self.resnet.eval()
+        if isinstance(X_test, pd.DataFrame):
+            X_np = X_test.values.astype(np.float32)
+        else:
+            X_np = X_test
+
+        with torch.no_grad():
+            y_pred = self(X_np).cpu().numpy()
+
+        if self.task_type == 'classification':
+            y_pred = 1 / (1 + np.exp(-y_pred))  # Sigmoid
+        else:
+            y_pred = np.clip(y_pred, 1e-6, None)
+        return y_pred.flatten()
+
+    # ---------------- 设置参数 ----------------
+
+    def set_params(self, params):
+        for key, value in params.items():
+            if hasattr(self, key):
+                setattr(self, key, value)
+            else:
+                raise ValueError(f"Parameter {key} not found in model.")
+        return self
+
+
+# =============================================================================
+# FT-Transformer model & sklearn-style wrapper
+# =============================================================================
+# 开始定义FT Transformer模型结构
+
+
+class FeatureTokenizer(nn.Module):
+    # 将数值与类别特征映射为 token，输出形状 (batch, token 数, d_model)
+    # 设定：
+    #   - X_num 表示数值特征，形状 (batch, num_numeric)
+    #   - X_cat 表示类别特征，形状 (batch, num_categorical)，每列为编码后的整数标签 [0, card-1]
+
+    def __init__(self, num_numeric: int, cat_cardinalities, d_model: int):
+        super().__init__()
+
+        self.num_numeric = num_numeric
+        self.has_numeric = num_numeric > 0
+
+        if self.has_numeric:
+            self.num_linear = nn.Linear(num_numeric, d_model)
+
+        self.embeddings = nn.ModuleList([
+            nn.Embedding(card, d_model) for card in cat_cardinalities
+        ])
+
+    def forward(self, X_num, X_cat):
+        tokens = []
+
+        if self.has_numeric:
+            # 数值特征映射为单个 token
+            num_token = self.num_linear(X_num)          # 形状 (batch, d_model)
+            tokens.append(num_token)
+
+        # 每个类别特征生成一个嵌入 token
+        for i, emb in enumerate(self.embeddings):
+            tok = emb(X_cat[:, i])                     # 形状 (batch, d_model)
+            tokens.append(tok)
+
+        # 最终堆叠为 (batch, token 数, d_model)
+        x = torch.stack(tokens, dim=1)
+        return x
+
+# 定义具有残差缩放的Encoder层
+
+
+class ScaledTransformerEncoderLayer(nn.Module):
+    def __init__(self, d_model: int, nhead: int, dim_feedforward: int = 2048,
+                 dropout: float = 0.1, residual_scale_attn: float = 1.0,
+                 residual_scale_ffn: float = 1.0, norm_first: bool = True,
+                 ):
+        super().__init__()
+        self.self_attn = nn.MultiheadAttention(
+            embed_dim=d_model,
+            num_heads=nhead,
+            dropout=dropout,
+            batch_first=True
+        )
+
+        # 前馈网络部分
+        self.linear1 = nn.Linear(d_model, dim_feedforward)
+        self.dropout = nn.Dropout(dropout)
+        self.linear2 = nn.Linear(dim_feedforward, d_model)
+
+        # 归一化与 Dropout
+        self.norm1 = nn.LayerNorm(d_model)
+        self.norm2 = nn.LayerNorm(d_model)
+        self.dropout1 = nn.Dropout(dropout)
+        self.dropout2 = nn.Dropout(dropout)
+
+        self.activation = nn.GELU()
+        # self.activation = nn.ReLU()
+        self.norm_first = norm_first
+
+        # 残差缩放系数
+        self.res_scale_attn = residual_scale_attn
+        self.res_scale_ffn = residual_scale_ffn
+
+    def forward(self, src, src_mask=None, src_key_padding_mask=None):
+        # 输入张量形状：(batch, 序列长度, d_model)
+        x = src
+
+        if self.norm_first:
+            # 先归一化再做注意力
+            x = x + self._sa_block(self.norm1(x), src_mask,
+                                   src_key_padding_mask)
+            x = x + self._ff_block(self.norm2(x))
+        else:
+            # 后归一化（一般不启用）
+            x = self.norm1(
+                x + self._sa_block(x, src_mask, src_key_padding_mask))
+            x = self.norm2(x + self._ff_block(x))
+
+        return x
+
+    def _sa_block(self, x, attn_mask, key_padding_mask):
+        # 自注意力并附带残差缩放
+        attn_out, _ = self.self_attn(
+            x, x, x,
+            attn_mask=attn_mask,
+            key_padding_mask=key_padding_mask,
+            need_weights=False
+        )
+        return self.res_scale_attn * self.dropout1(attn_out)
+
+    def _ff_block(self, x):
+        # 前馈网络并附带残差缩放
+        x2 = self.linear2(self.dropout(self.activation(self.linear1(x))))
+        return self.res_scale_ffn * self.dropout2(x2)
+
+# 定义FT-Transformer核心模型
+
+
+class FTTransformerCore(nn.Module):
+    # 最小可用版本的 FT-Transformer：
+    #   - FeatureTokenizer：将数值与类别特征转换为 token
+    #   - TransformerEncoder：捕捉特征之间的交互
+    #   - 池化 + MLP + Softplus：保证输出为正值（适配 Tweedie/Gamma）
+
+    def __init__(self, num_numeric: int, cat_cardinalities, d_model: int = 64,
+                 n_heads: int = 8, n_layers: int = 4, dropout: float = 0.1,
+                 ):
+        super().__init__()
+
+        self.tokenizer = FeatureTokenizer(
+            num_numeric=num_numeric,
+            cat_cardinalities=cat_cardinalities,
+            d_model=d_model
+        )
+        scale = 1.0 / math.sqrt(n_layers)  # 推荐一个默认值
+        encoder_layer = ScaledTransformerEncoderLayer(
+            d_model=d_model,
+            nhead=n_heads,
+            dim_feedforward=d_model * 4,
+            dropout=dropout,
+            residual_scale_attn=scale,
+            residual_scale_ffn=scale,
+            norm_first=True,
+        )
+        self.encoder = nn.TransformerEncoder(
+            encoder_layer,
+            num_layers=n_layers
+        )
+        self.n_layers = n_layers
+
+        self.head = nn.Sequential(
+            nn.LayerNorm(d_model),
+            nn.Linear(d_model, d_model),
+            nn.GELU(),
+            # nn.ReLU(),
+            nn.Linear(d_model, 1),
+            # nn.Softplus()  # 保证输出为正,适合 Tweedie / Gamma
+            # 移除 Softplus，让模型输出 logits。
+            # 在训练和推理时根据任务类型决定是否应用 sigmoid 或 softplus。
+            # 对于分类，输出 logits，然后用 BCEWithLogitsLoss
+            # 对于回归，在推理时应用 softplus
+        )
+
+    def forward(self, X_num, X_cat):
+
+        # X_num: (batch, 数值特征数)，float32
+        # X_cat: (batch, 类别特征数)，long
+
+        tokens = self.tokenizer(X_num, X_cat)  # 形状 (batch, token 数, d_model)
+        x = self.encoder(tokens)               # 形状 (batch, token 数, d_model)
+
+        # 对 token 做平均池化
+        x = x.mean(dim=1)                      # 形状 (batch, d_model)
+
+        out = self.head(x)                     # 形状 (batch, 1)，Softplus 保证为正
+        return out
+
+# 定义TabularDataset类
+
+
+class TabularDataset(Dataset):
+    def __init__(self, X_num, X_cat, y, w):
+
+        # X_num: torch.float32, 形状 (N, 数值特征数)
+        # X_cat: torch.long,    形状 (N, 类别特征数)
+        # y:     torch.float32, 形状 (N, 1)
+        # w:     torch.float32, 形状 (N, 1)
+
+        self.X_num = X_num
+        self.X_cat = X_cat
+        self.y = y
+        self.w = w
+
+    def __len__(self):
+        return self.y.shape[0]
+
+    def __getitem__(self, idx):
+        return (
+            self.X_num[idx],
+            self.X_cat[idx],
+            self.y[idx],
+            self.w[idx],
+        )
+
+# 定义FTTransformer的Scikit-Learn接口类
+
+
+class FTTransformerSklearn(nn.Module):
+
+    # sklearn 风格包装：
+    #   - num_cols：数值特征列名列表
+    #   - cat_cols：类别特征列名列表（需提前做标签编码，取值 [0, n_classes-1]）
+
+    def __init__(self, model_nme: str, num_cols, cat_cols, d_model: int = 64, n_heads: int = 8,
+                 n_layers: int = 4, dropout: float = 0.1, batch_num: int = 100, epochs: int = 100,
+                 task_type: str = 'regression',
+                 tweedie_power: float = 1.5, learning_rate: float = 1e-3, patience: int = 10,
+                 use_data_parallel: bool = True,
+                 ):
+        super().__init__()
+
+        self.model_nme = model_nme
+        self.num_cols = list(num_cols)
+        self.cat_cols = list(cat_cols)
+        self.d_model = d_model
+        self.n_heads = n_heads
+        self.n_layers = n_layers
+        self.dropout = dropout
+        self.batch_num = batch_num
+        self.epochs = epochs
+        self.learning_rate = learning_rate
+        self.task_type = task_type
+        self.patience = patience
+        if 'f' in self.model_nme:
+            self.tw_power = 1.0
+        elif 's' in self.model_nme:
+            self.tw_power = 2.0
+        else:
+            self.tw_power = tweedie_power
+        if torch.cuda.is_available():
+            self.device = torch.device("cuda")
+        elif torch.backends.mps.is_available():
+            self.device = torch.device("mps")
+        else:
+            self.device = torch.device("cpu")
+        self.cat_cardinalities = None
+        self.cat_categories = {}
+        self.ft = None
+        self.use_data_parallel = torch.cuda.device_count() > 1 and use_data_parallel
+
+    def _build_model(self, X_train):
+        num_numeric = len(self.num_cols)
+        cat_cardinalities = []
+
+        for col in self.cat_cols:
+            cats = X_train[col].astype('category')
+            categories = cats.cat.categories
+            self.cat_categories[col] = categories           # 保存训练集类别全集
+
+            card = len(categories) + 1                      # 多预留 1 类给“未知/缺失”
+            cat_cardinalities.append(card)
+
+        self.cat_cardinalities = cat_cardinalities
+
+        core = FTTransformerCore(
+            num_numeric=num_numeric,
+            cat_cardinalities=cat_cardinalities,
+            d_model=self.d_model,
+            n_heads=self.n_heads,
+            n_layers=self.n_layers,
+            dropout=self.dropout,
+        )
+        if self.use_data_parallel:
+            core = nn.DataParallel(core, device_ids=list(range(torch.cuda.device_count())))
+            self.device = torch.device("cuda")
+        self.ft = core.to(self.device)
+
+    def _encode_cats(self, X):
+        # 输入 DataFrame 至少需要包含所有类别特征列
+        # 返回形状 (N, 类别特征数) 的 int64 数组
+
+        if not self.cat_cols:
+            return np.zeros((len(X), 0), dtype='int64')
+
+        X_cat_list = []
+        for col in self.cat_cols:
+            # 使用训练阶段记录的类别全集
+            categories = self.cat_categories[col]
+            # 按固定类别构造 Categorical
+            cats = pd.Categorical(X[col], categories=categories)
+            codes = cats.codes.astype('int64', copy=True)   # -1 表示未知或缺失
+            # 未知或缺失映射到额外的“未知”索引 len(categories)
+            codes[codes < 0] = len(categories)
+            X_cat_list.append(codes)
+
+        X_cat_np = np.stack(X_cat_list, axis=1)  # 形状 (N, 类别特征数)
+        return X_cat_np
+
+    def fit(self, X_train, y_train, w_train=None,
+            X_val=None, y_val=None, w_val=None):
+    
+        # 首次拟合时需要构建底层模型结构
+        if self.ft is None:
+            self._build_model(X_train)
+    
+        # --- 构建训练张量（全部先放在 CPU，后续按批搬运） ---
+        # 数值特征
+        X_num_train = X_train[self.num_cols].to_numpy(
+            dtype=np.float32, copy=True
+        )
+        X_num_train = torch.tensor(
+            X_num_train,
+            dtype=torch.float32
+        )
+    
+        # 类别特征
+        if self.cat_cols:
+            X_cat_train_np = self._encode_cats(X_train)
+            X_cat_train = torch.tensor(X_cat_train_np, dtype=torch.long)
+        else:
+            X_cat_train = torch.zeros(
+                (X_num_train.shape[0], 0), dtype=torch.long
+            )
+    
+        # 目标 & 权重
+        y_tensor = torch.tensor(
+            y_train.values,
+            dtype=torch.float32
+        ).view(-1, 1)
+    
+        if w_train is not None:
+            w_tensor = torch.tensor(
+                w_train.values,
+                dtype=torch.float32
+            ).view(-1, 1)
+        else:
+            w_tensor = torch.ones_like(y_tensor)
+    
+        # --- 验证集张量（一次性搬到目标设备） ---
+        has_val = X_val is not None and y_val is not None
+        if has_val:
+            # 数值特征
+            X_num_val_np = X_val[self.num_cols].to_numpy(
+                dtype=np.float32, copy=True
+            )
+            X_num_val = torch.tensor(X_num_val_np, dtype=torch.float32)
+    
+            # 类别特征
+            if self.cat_cols:
+                X_cat_val_np = self._encode_cats(X_val)
+                X_cat_val = torch.tensor(X_cat_val_np, dtype=torch.long)
+            else:
+                X_cat_val = torch.zeros(
+                    (X_num_val.shape[0], 0), dtype=torch.long
+                )
+    
+            # 目标 & 权重
+            y_val_np = y_val.values.astype(np.float32, copy=True)
+            y_val_tensor = torch.tensor(
+                y_val_np, dtype=torch.float32
+            ).view(-1, 1)
+    
+            if w_val is not None:
+                w_val_np = w_val.values.astype(np.float32, copy=True)
+                w_val_tensor = torch.tensor(
+                    w_val_np, dtype=torch.float32
+                ).view(-1, 1)
+            else:
+                w_val_tensor = torch.ones_like(y_val_tensor)
+        else:
+            X_num_val = X_cat_val = y_val_tensor = w_val_tensor = None
+    
+        # --- 构建 DataLoader ---
+        dataset = TabularDataset(
+            X_num_train, X_cat_train, y_tensor, w_tensor
+        )
+    
+        # 原来的 batch 估算逻辑：先给一个初始 batch_size
+        batch_size = compute_batch_size(
+            data_size=len(dataset),
+            learning_rate=self.learning_rate,
+            batch_num=self.batch_num,
+            minimum=64
+        )
+    
+        N = X_num_train.shape[0]
+        if self.device.type == 'cuda':
+            if N > 200_000:
+                base_bs = 512
+            elif N > 50_000:
+                base_bs = 256
+            else:
+                base_bs = 128
+        else:
+            if N > 50_000:
+                base_bs = 256
+            else:
+                base_bs = 128
+    
+        # 真实的物理 batch（显存安全为主）
+        batch_size = min(batch_size, base_bs, N)
+    
+        # ====== ⭐ 新增：梯度累积设置 ======
+        # 目标“有效 batch 大小”（影响收敛稳定性）
+        # 可以根据经验稍微调大/调小，比如 2048 / 4096 / 8192
+        target_effective_bs = 4096 if self.device.type == 'cuda' else 2048
+        accum_steps = max(1, target_effective_bs // batch_size)
+    
+        dataloader = DataLoader(
+            dataset,
+            batch_size=batch_size,
+            shuffle=True,
+            num_workers=1,
+            pin_memory=(self.device.type == 'cuda')
+        )
+    
+        # --- 优化器与 AMP ---
+        optimizer = torch.optim.Adam(
+            self.ft.parameters(),
+            lr=self.learning_rate
+        )
+        scaler = GradScaler(enabled=(self.device.type == 'cuda'))
+    
+        # --- 早停机制 ---
+        best_loss = float('inf')
+        patience_counter = 0
+        best_model_state = None
+    
+        # 若存在验证集则整体迁移到目标设备
+        if has_val:
+            X_num_val_dev = X_num_val.to(self.device, non_blocking=True)
+            X_cat_val_dev = X_cat_val.to(self.device, non_blocking=True)
+            y_val_dev = y_val_tensor.to(self.device, non_blocking=True)
+            w_val_dev = w_val_tensor.to(self.device, non_blocking=True)
+    
+        # --- 训练循环 ---
+        for epoch in range(1, self.epochs + 1):
+            self.ft.train()
+            optimizer.zero_grad()  # ⭐ 梯度在 epoch 内累积，所以放在这里
+    
+            for step, (X_num_b, X_cat_b, y_b, w_b) in enumerate(dataloader):
+    
+                X_num_b = X_num_b.to(self.device, non_blocking=True)
+                X_cat_b = X_cat_b.to(self.device, non_blocking=True)
+                y_b = y_b.to(self.device, non_blocking=True)
+                w_b = w_b.to(self.device, non_blocking=True)
+    
+                with autocast(enabled=(self.device.type == 'cuda')):
+                    y_pred = self.ft(X_num_b, X_cat_b)
+    
+                    if self.task_type == 'classification':
+                        loss_fn = nn.BCEWithLogitsLoss(reduction='none')
+                        losses = loss_fn(y_pred, y_b).view(-1)
+                    else:
+                        # 对于回归，需要保证预测值为正
+                        y_pred = F.softplus(y_pred)
+                        y_pred = torch.clamp(y_pred, min=1e-6)
+                        losses = tweedie_loss(
+                            y_pred, y_b, p=self.tw_power
+                        ).view(-1)
+    
+                    weighted_loss = (
+                        losses * w_b.view(-1)
+                    ).sum() / torch.clamp(w_b.sum(), min=EPS)
+    
+                    # ⭐ 梯度累积：把 loss 平均到 accum_steps
+                    loss_for_backward = weighted_loss / accum_steps
+    
+                scaler.scale(loss_for_backward).backward()
+    
+                # ⭐ 到了累积步数或最后一个 batch，才真正 step 一次
+                if ((step + 1) % accum_steps == 0) or ((step + 1) == len(dataloader)):
+                    # 可选：梯度裁剪
+                    if self.device.type == 'cuda':
+                        scaler.unscale_(optimizer)
+                        clip_grad_norm_(self.ft.parameters(), max_norm=1.0)
+
+                    scaler.step(optimizer)
+                    scaler.update()
+                    optimizer.zero_grad()
+    
+            # --- 验证阶段与早停判断 ---
+            if has_val:
+                self.ft.eval()
+                with torch.no_grad(), autocast(enabled=(self.device.type == 'cuda')):
+                    y_val_pred = self.ft(X_num_val_dev, X_cat_val_dev)
+    
+                    if self.task_type == 'classification':
+                        val_loss_fn = nn.BCEWithLogitsLoss(reduction='none')
+                        val_losses = val_loss_fn(
+                            y_val_pred, y_val_dev
+                        ).view(-1)
+                    else:
+                        y_val_pred = F.softplus(y_val_pred)
+                        y_val_pred = torch.clamp(y_val_pred, min=1e-6)
+                        val_losses = tweedie_loss(
+                            y_val_pred, y_val_dev, p=self.tw_power
+                        ).view(-1)
+    
+                    val_weighted_loss = (
+                        val_losses * w_val_dev.view(-1)
+                    ).sum() / torch.clamp(w_val_dev.sum(), min=EPS)
+    
+                if val_weighted_loss < best_loss:
+                    best_loss = val_weighted_loss
+                    patience_counter = 0
+                    best_model_state = copy.deepcopy(self.ft.state_dict())
+                else:
+                    patience_counter += 1
+    
+                if patience_counter >= self.patience and best_model_state is not None:
+                    self.ft.load_state_dict(best_model_state)
+                    break
+    
+        if has_val and best_model_state is not None:
+            self.ft.load_state_dict(best_model_state)
+
+
+    def predict(self, X_test):
+        # X_test 需要包含所有数值列与类别列
+
+        self.ft.eval()
+        X_num = X_test[self.num_cols].to_numpy(dtype=np.float32, copy=True)
+        X_num = torch.tensor(
+            X_num,
+            dtype=torch.float32
+        )
+        if self.cat_cols:
+            X_cat_np = self._encode_cats(X_test)
+            X_cat = torch.tensor(X_cat_np, dtype=torch.long)
+        else:
+            X_cat = torch.zeros((X_num.size(0), 0), dtype=torch.long)
+
+        with torch.no_grad():
+            X_num = X_num.to(self.device, non_blocking=True)
+            X_cat = X_cat.to(self.device, non_blocking=True)
+            y_pred = self.ft(X_num, X_cat).cpu().numpy()
+
+        if self.task_type == 'classification':
+            # 从 logits 转换为概率
+            y_pred = 1 / (1 + np.exp(-y_pred))
+        else:
+            y_pred = np.log(1 + np.exp(y_pred))  # softplus
+            y_pred = np.clip(y_pred, 1e-6, None)
+        return y_pred.ravel()
+
+    def set_params(self, params: dict):
+
+        # 和 sklearn 风格保持一致。
+        # 注意：对结构性参数（如 d_model/n_heads）修改后,需要重新 fit 才会生效。
+
+        for key, value in params.items():
+            if hasattr(self, key):
+                setattr(self, key, value)
+            else:
+                raise ValueError(f"Parameter {key} not found in model.")
+        return self
+
+
+# ===== 基础组件与训练封装 =====================================================
+
+# =============================================================================
+# Config, preprocessing, and trainer base
+# =============================================================================
+@dataclass
+class BayesOptConfig:
+    model_nme: str
+    resp_nme: str
+    weight_nme: str
+    factor_nmes: List[str]
+    task_type: str = 'regression'
+    binary_resp_nme: Optional[str] = None
+    cate_list: Optional[List[str]] = None
+    prop_test: float = 0.25
+    rand_seed: Optional[int] = None
+    epochs: int = 100
+    use_gpu: bool = True
+    use_resn_data_parallel: bool = True
+    use_ft_data_parallel: bool = True
+
+
+class OutputManager:
+    # 统一管理结果、图表与模型的输出路径
+
+    def __init__(self, root: Optional[str] = None, model_name: str = "model") -> None:
+        self.root = Path(root or os.getcwd())
+        self.model_name = model_name
+        self.plot_dir = self.root / 'plot'
+        self.result_dir = self.root / 'Results'
+        self.model_dir = self.root / 'model'
+
+    def _prepare(self, path: Path) -> str:
+        ensure_parent_dir(str(path))
+        return str(path)
+
+    def plot_path(self, filename: str) -> str:
+        return self._prepare(self.plot_dir / filename)
+
+    def result_path(self, filename: str) -> str:
+        return self._prepare(self.result_dir / filename)
+
+    def model_path(self, filename: str) -> str:
+        return self._prepare(self.model_dir / filename)
+
+
+class DatasetPreprocessor:
+    # 为各训练器准备通用的训练/测试数据视图
+
+    def __init__(self, train_df: pd.DataFrame, test_df: pd.DataFrame,
+                 config: BayesOptConfig) -> None:
+        self.config = config
+        self.train_data = train_df.copy(deep=True)
+        self.test_data = test_df.copy(deep=True)
+        self.num_features: List[str] = []
+        self.train_oht_scl_data: Optional[pd.DataFrame] = None
+        self.test_oht_scl_data: Optional[pd.DataFrame] = None
+        self.var_nmes: List[str] = []
+        self.cat_categories_for_shap: Dict[str, List[Any]] = {}
+
+    def run(self) -> "DatasetPreprocessor":
+        cfg = self.config
+        # 预先计算加权实际值，后续画图、校验都依赖该字段
+        self.train_data.loc[:, 'w_act'] = self.train_data[cfg.resp_nme] * \
+            self.train_data[cfg.weight_nme]
+        self.test_data.loc[:, 'w_act'] = self.test_data[cfg.resp_nme] * \
+            self.test_data[cfg.weight_nme]
+        if cfg.binary_resp_nme:
+            self.train_data.loc[:, 'w_binary_act'] = self.train_data[cfg.binary_resp_nme] * \
+                self.train_data[cfg.weight_nme]
+            self.test_data.loc[:, 'w_binary_act'] = self.test_data[cfg.binary_resp_nme] * \
+                self.test_data[cfg.weight_nme]
+        # 高分位裁剪用来吸收离群值；若删除会导致极端点主导损失
+        q99 = self.train_data[cfg.resp_nme].quantile(0.999)
+        self.train_data[cfg.resp_nme] = self.train_data[cfg.resp_nme].clip(
+            upper=q99)
+        cate_list = list(cfg.cate_list or [])
+        if cate_list:
+            for cate in cate_list:
+                self.train_data[cate] = self.train_data[cate].astype(
+                    'category')
+                self.test_data[cate] = self.test_data[cate].astype('category')
+                cats = self.train_data[cate].cat.categories
+                self.cat_categories_for_shap[cate] = list(cats)
+        self.num_features = [
+            nme for nme in cfg.factor_nmes if nme not in cate_list]
+        train_oht = self.train_data[cfg.factor_nmes +
+                                    [cfg.weight_nme] + [cfg.resp_nme]].copy()
+        test_oht = self.test_data[cfg.factor_nmes +
+                                  [cfg.weight_nme] + [cfg.resp_nme]].copy()
+        train_oht = pd.get_dummies(
+            train_oht,
+            columns=cate_list,
+            drop_first=True,
+            dtype=np.int8
+        )
+        test_oht = pd.get_dummies(
+            test_oht,
+            columns=cate_list,
+            drop_first=True,
+            dtype=np.int8
+        )
+        for num_chr in self.num_features:
+            # 逐列标准化保障每个特征在同一量级，否则神经网络会难以收敛
+            scaler = StandardScaler()
+            train_oht[num_chr] = scaler.fit_transform(
+                train_oht[num_chr].values.reshape(-1, 1))
+            test_oht[num_chr] = scaler.transform(
+                test_oht[num_chr].values.reshape(-1, 1))
+        # reindex 时将缺失的哑变量列补零，避免测试集列数与训练集不一致
+        test_oht = test_oht.reindex(columns=train_oht.columns, fill_value=0)
+        self.train_oht_scl_data = train_oht
+        self.test_oht_scl_data = test_oht
+        self.var_nmes = list(
+            set(list(train_oht.columns)) - set([cfg.weight_nme, cfg.resp_nme])
+        )
+        return self
+
+# =============================================================================
+# Trainers
+# =============================================================================
+
+
+class TrainerBase:
+    def __init__(self, context: "BayesOptModel", label: str) -> None:
+        self.ctx = context
+        self.label = label
+
+    @property
+    def config(self) -> BayesOptConfig:
+        return self.ctx.config
+
+    @property
+    def output(self) -> OutputManager:
+        return self.ctx.output_manager
+
+    def tune(self, max_evals: int) -> None:  # pragma: no cover 子类会覆盖
+        raise NotImplementedError
+
+    def train(self) -> None:  # pragma: no cover 子类会覆盖
+        raise NotImplementedError
+
+    def save(self) -> None:
+        pass
+
+    def load(self) -> None:
+        pass
+
+
+class XGBTrainer(TrainerBase):
+    def __init__(self, context: "BayesOptModel") -> None:
+        super().__init__(context, 'Xgboost')
+        self.model: Optional[xgb.XGBRegressor] = None
+        self.best_params: Optional[Dict[str, Any]] = None
+        self.best_trial = None
+
+    def _build_estimator(self) -> xgb.XGBRegressor:
+        params = dict(
+            objective=self.ctx.obj,
+            random_state=self.ctx.rand_seed,
+            subsample=0.9,
+            tree_method='gpu_hist' if self.ctx.use_gpu else 'hist',
+            enable_categorical=True,
+            predictor='gpu_predictor' if self.ctx.use_gpu else 'cpu_predictor'
+        )
+        if self.ctx.use_gpu:
+            params['gpu_id'] = 0
+        return xgb.XGBRegressor(**params)
+
+    def cross_val(self, trial: optuna.trial.Trial) -> float:
+        learning_rate = trial.suggest_float(
+            'learning_rate', 1e-5, 1e-1, log=True)
+        gamma = trial.suggest_float('gamma', 0, 10000)
+        max_depth = trial.suggest_int('max_depth', 3, 25)
+        n_estimators = trial.suggest_int('n_estimators', 10, 500, step=10)
+        min_child_weight = trial.suggest_int(
+            'min_child_weight', 100, 10000, step=100)
+        reg_alpha = trial.suggest_float('reg_alpha', 1e-10, 1, log=True)
+        reg_lambda = trial.suggest_float('reg_lambda', 1e-10, 1, log=True)
+        if self.ctx.obj == 'reg:tweedie':
+            tweedie_variance_power = trial.suggest_float(
+                'tweedie_variance_power', 1, 2)
+        elif self.ctx.obj == 'count:poisson':
+            tweedie_variance_power = 1
+        elif self.ctx.obj == 'reg:gamma':
+            tweedie_variance_power = 2
+        else:
+            tweedie_variance_power = 1.5
+        clf = self._build_estimator()
+        params = {
+            'learning_rate': learning_rate,
+            'gamma': gamma,
+            'max_depth': max_depth,
+            'n_estimators': n_estimators,
+            'min_child_weight': min_child_weight,
+            'reg_alpha': reg_alpha,
+            'reg_lambda': reg_lambda
+        }
+        if self.ctx.obj == 'reg:tweedie':
+            params['tweedie_variance_power'] = tweedie_variance_power
+        clf.set_params(**params)
+        n_jobs = 1 if self.ctx.use_gpu else int(1 / self.ctx.prop_test)
+        acc = cross_val_score(
+            clf,
+            self.ctx.train_data[self.ctx.factor_nmes],
+            self.ctx.train_data[self.ctx.resp_nme].values,
+            fit_params=self.ctx.fit_params,
+            cv=self.ctx.cv,
+            scoring=make_scorer(
+                mean_tweedie_deviance,
+                power=tweedie_variance_power,
+                greater_is_better=False),
+            error_score='raise',
+            n_jobs=n_jobs
+        ).mean()
+        return -acc
+
+    def tune(self, max_evals: int = 100) -> None:
+        study = optuna.create_study(
+            direction='minimize',
+            sampler=optuna.samplers.TPESampler(seed=self.ctx.rand_seed)
+        )
+        study.optimize(self.cross_val, n_trials=max_evals)
+        self.best_params = study.best_params
+        self.best_trial = study.best_trial
+        params_path = self.output.result_path(
+            f'{self.ctx.model_nme}_bestparams_xgb.csv'
+        )
+        pd.DataFrame(self.best_params, index=[0]).to_csv(params_path)
+
+    def train(self) -> None:
+        if not self.best_params:
+            raise RuntimeError('请先运行 tune() 以获得 XGB 最优参数。')
+        self.model = self._build_estimator()
+        self.model.set_params(**self.best_params)
+        self.model.fit(self.ctx.train_data[self.ctx.factor_nmes],
+                       self.ctx.train_data[self.ctx.resp_nme].values,
+                       **self.ctx.fit_params)
+        self.ctx.model_label += [self.label]
+        self.ctx.train_data['pred_xgb'] = self.model.predict(
+            self.ctx.train_data[self.ctx.factor_nmes])
+        self.ctx.test_data['pred_xgb'] = self.model.predict(
+            self.ctx.test_data[self.ctx.factor_nmes])
+        self.ctx.train_data.loc[:, 'w_pred_xgb'] = self.ctx.train_data['pred_xgb'] * \
+            self.ctx.train_data[self.ctx.weight_nme]
+        self.ctx.test_data.loc[:, 'w_pred_xgb'] = self.ctx.test_data['pred_xgb'] * \
+            self.ctx.test_data[self.ctx.weight_nme]
+        self.ctx.xgb_best = self.model
+
+    def save(self) -> None:
+        if self.model is not None:
+            joblib.dump(self.model, self.output.model_path(
+                f'01_{self.ctx.model_nme}_Xgboost.pkl'))
+
+    def load(self) -> None:
+        path = self.output.model_path(
+            f'01_{self.ctx.model_nme}_Xgboost.pkl')
+        if os.path.exists(path):
+            self.model = joblib.load(path)
+            self.ctx.xgb_best = self.model
+        else:
+            print(f"[load_model] Warning: 未找到 Xgboost 模型文件：{path}")
+
+
+class GLMTrainer(TrainerBase):
+    def __init__(self, context: "BayesOptModel") -> None:
+        super().__init__(context, 'GLM')
+        self.model = None
+        self.best_params: Optional[Dict[str, Any]] = None
+        self.best_trial = None
+
+    def _select_family(self, tweedie_power: Optional[float] = None):
+        if self.ctx.task_type == 'classification':
+            return sm.families.Binomial()
+        if self.ctx.obj == 'count:poisson':
+            return sm.families.Poisson()
+        if self.ctx.obj == 'reg:gamma':
+            return sm.families.Gamma()
+        power = tweedie_power if tweedie_power is not None else 1.5
+        return sm.families.Tweedie(var_power=power, link=sm.families.links.log())
+
+    def _prepare_design(self, data: pd.DataFrame) -> pd.DataFrame:
+        # 为 statsmodels 添加截距项
+        X = data[self.ctx.var_nmes]
+        return sm.add_constant(X, has_constant='add')
+
+    def _metric_power(self, family, tweedie_power: Optional[float]) -> float:
+        if isinstance(family, sm.families.Poisson):
+            return 1.0
+        if isinstance(family, sm.families.Gamma):
+            return 2.0
+        if isinstance(family, sm.families.Tweedie):
+            return tweedie_power if tweedie_power is not None else getattr(family, 'var_power', 1.5)
+        return 1.5
+
+    def cross_val(self, trial: optuna.trial.Trial) -> float:
+        alpha = trial.suggest_float('alpha', 1e-6, 1e2, log=True)
+        l1_ratio = trial.suggest_float('l1_ratio', 0.0, 1.0)
+        tweedie_power = None
+        if self.ctx.task_type == 'regression' and self.ctx.obj == 'reg:tweedie':
+            tweedie_power = trial.suggest_float('tweedie_power', 1.01, 1.99)
+
+        X_all = self._prepare_design(self.ctx.train_oht_scl_data)
+        y_all = self.ctx.train_oht_scl_data[self.ctx.resp_nme]
+        w_all = self.ctx.train_oht_scl_data[self.ctx.weight_nme]
+
+        scores = []
+        for train_idx, val_idx in self.ctx.cv.split(X_all):
+            X_train, X_val = X_all.iloc[train_idx], X_all.iloc[val_idx]
+            y_train, y_val = y_all.iloc[train_idx], y_all.iloc[val_idx]
+            w_train, w_val = w_all.iloc[train_idx], w_all.iloc[val_idx]
+
+            family = self._select_family(tweedie_power)
+            glm = sm.GLM(y_train, X_train, family=family,
+                         freq_weights=w_train)
+            result = glm.fit_regularized(
+                alpha=alpha, L1_wt=l1_ratio, maxiter=200)
+
+            y_pred = result.predict(X_val)
+            if self.ctx.task_type == 'classification':
+                y_pred = np.clip(y_pred, EPS, 1 - EPS)
+                fold_score = log_loss(
+                    y_val, y_pred, sample_weight=w_val)
+            else:
+                y_pred = np.maximum(y_pred, EPS)
+                fold_score = mean_tweedie_deviance(
+                    y_val,
+                    y_pred,
+                    sample_weight=w_val,
+                    power=self._metric_power(family, tweedie_power)
+                )
+            scores.append(fold_score)
+
+        return float(np.mean(scores))
+
+    def tune(self, max_evals: int = 50) -> None:
+        study = optuna.create_study(
+            direction='minimize', sampler=optuna.samplers.TPESampler(seed=self.ctx.rand_seed))
+        study.optimize(self.cross_val, n_trials=max_evals)
+        self.best_params = study.best_params
+        self.best_trial = study.best_trial
+        params_path = self.output.result_path(
+            f'{self.ctx.model_nme}_bestparams_glm.csv')
+        pd.DataFrame(self.best_params, index=[0]).to_csv(params_path)
+
+    def train(self) -> None:
+        if not self.best_params:
+            raise RuntimeError('请先运行 tune() 以获得 GLM 最优参数。')
+        tweedie_power = self.best_params.get('tweedie_power')
+        family = self._select_family(tweedie_power)
+
+        X_train = self._prepare_design(self.ctx.train_oht_scl_data)
+        y_train = self.ctx.train_oht_scl_data[self.ctx.resp_nme]
+        w_train = self.ctx.train_oht_scl_data[self.ctx.weight_nme]
+
+        glm = sm.GLM(y_train, X_train, family=family,
+                     freq_weights=w_train)
+        self.model = glm.fit_regularized(
+            alpha=self.best_params['alpha'],
+            L1_wt=self.best_params['l1_ratio'],
+            maxiter=300
+        )
+
+        self.ctx.glm_best = self.model
+        self.ctx.model_label += [self.label]
+        self._attach_predictions(tweedie_power)
+
+    def _attach_predictions(self, tweedie_power: Optional[float]) -> None:
+        for src, target in (
+            (self.ctx.train_oht_scl_data, self.ctx.train_data),
+            (self.ctx.test_oht_scl_data, self.ctx.test_data),
+        ):
+            design = self._prepare_design(src)
+            preds = self.model.predict(design)
+            if self.ctx.task_type == 'classification':
+                preds = np.clip(preds, EPS, 1 - EPS)
+            else:
+                preds = np.maximum(preds, EPS)
+            target['pred_glm'] = preds
+            target['w_pred_glm'] = target['pred_glm'] * \
+                target[self.ctx.weight_nme]
+
+    def save(self) -> None:
+        if self.model is not None:
+            joblib.dump(self.model, self.output.model_path(
+                f'01_{self.ctx.model_nme}_GLM.pkl'))
+
+    def load(self) -> None:
+        path = self.output.model_path(
+            f'01_{self.ctx.model_nme}_GLM.pkl')
+        if os.path.exists(path):
+            self.model = joblib.load(path)
+            self.ctx.glm_best = self.model
+        else:
+            print(f"[load_model] Warning: 未找到 GLM 模型文件：{path}")
+
+
+class ResNetTrainer(TrainerBase):
+    def __init__(self, context: "BayesOptModel") -> None:
+        if context.task_type == 'classification':
+            super().__init__(context, 'ResNetClassifier')
+        else:
+            super().__init__(context, 'ResNet')
+        self.model: Optional[ResNetSklearn] = None
+        self.best_params: Optional[Dict[str, Any]] = None
+        self.best_trial = None
+
+    # ========= 交叉验证（BayesOpt 用） =========
+    def cross_val(self, trial: optuna.trial.Trial) -> float:
+        """
+        对 ResNet 做交叉验证。
+        为了防止显存 OOM：
+          - 每个 fold 独立创建一个 ResNetSklearn
+          - fold 结束就把模型挪到 CPU + 删除 + gc + empty_cache
+          - 可选：BayesOpt 阶段只用训练集子样本
+        """
+
+        # 1. 超参空间（基本沿用你之前的设定）
+        learning_rate = trial.suggest_float(
+            'learning_rate', 1e-6, 1e-2, log=True
+        )
+        # hidden_dim = trial.suggest_int('hidden_dim', 32, 256, step=32) # 不宜过大
+        hidden_dim = trial.suggest_int('hidden_dim', 8, 32, step=2)
+        block_num = trial.suggest_int('block_num', 2, 10)
+        # batch_num = trial.suggest_int(
+        #     'batch_num',
+        #     10 if self.ctx.obj == 'reg:gamma' else 100,
+        #     100 if self.ctx.obj == 'reg:gamma' else 1000,
+        #     step=10 if self.ctx.obj == 'reg:gamma' else 100
+        # )
+
+        if self.ctx.task_type == 'regression':
+            if self.ctx.obj == 'reg:tweedie':
+                tw_power = trial.suggest_float('tw_power', 1.0, 2.0)
+            elif self.ctx.obj == 'count:poisson':
+                tw_power = 1.0
+            elif self.ctx.obj == 'reg:gamma':
+                tw_power = 2.0
+            else:
+                tw_power = 1.5
+        else:  # classification
+            tw_power = None  # Not used
+
+        fold_losses = []
+
+        # 2. （可选）BayesOpt 只在子样本上做 CV，减轻显存 & 时间压力
+        data_for_cv = self.ctx.train_oht_scl_data
+        max_rows_for_resnet_bo = min(100000, int(
+            len(data_for_cv)/5))  # 你可以按 A30 情况调小，比如 50_000
+        if len(data_for_cv) > max_rows_for_resnet_bo:
+            data_for_cv = data_for_cv.sample(
+                max_rows_for_resnet_bo,
+                random_state=self.ctx.rand_seed
+            )
+
+        X_all = data_for_cv[self.ctx.var_nmes]
+        y_all = data_for_cv[self.ctx.resp_nme]
+        w_all = data_for_cv[self.ctx.weight_nme]
+
+        # 用局部 ShuffleSplit，避免子样本时索引不一致
+        cv_local = ShuffleSplit(
+            n_splits=int(1 / self.ctx.prop_test),
+            test_size=self.ctx.prop_test,
+            random_state=self.ctx.rand_seed
+        )
+
+        for fold, (train_idx, val_idx) in enumerate(cv_local.split(X_all)):
+            X_train_fold = X_all.iloc[train_idx]
+            y_train_fold = y_all.iloc[train_idx]
+            w_train_fold = w_all.iloc[train_idx]
+
+            X_val_fold = X_all.iloc[val_idx]
+            y_val_fold = y_all.iloc[val_idx]
+            w_val_fold = w_all.iloc[val_idx]
+
+            # 3. 每个 fold 创建一个临时 ResNet 模型
+            cv_net = ResNetSklearn(
+                model_nme=self.ctx.model_nme,
+                input_dim=X_all.shape[1],
+                hidden_dim=hidden_dim,
+                block_num=block_num,
+                task_type=self.ctx.task_type,
+                # batch_num=batch_num,
+                epochs=self.ctx.epochs,
+                tweedie_power=tw_power,
+                learning_rate=learning_rate,
+                patience=5,
+                use_layernorm=True,
+                dropout=0.1,
+                residual_scale=0.1,
+                use_data_parallel=self.ctx.config.use_resn_data_parallel
+            )
+
+            try:
+                # 4. 训练（内部仍然用你自己的 tweedie_loss）
+                cv_net.fit(
+                    X_train_fold,
+                    y_train_fold,
+                    w_train_fold,
+                    X_val_fold,
+                    y_val_fold,
+                    w_val_fold
+                )
+
+                # 5. 验证集预测
+                y_pred_fold = cv_net.predict(X_val_fold)
+
+                # 6. 评估：Tweedie deviance（评估用，训练 loss 不动）
+                if self.ctx.task_type == 'regression':
+                    loss = mean_tweedie_deviance(
+                        y_val_fold,
+                        y_pred_fold,
+                        sample_weight=w_val_fold,
+                        power=tw_power
+                    )
+                else:  # classification
+                    from sklearn.metrics import log_loss
+                    loss = log_loss(
+                        y_val_fold,
+                        y_pred_fold,
+                        sample_weight=w_val_fold,
+                    )
+                fold_losses.append(loss)
+            finally:
+                # 7. ★ 每个 fold 结束后释放 GPU 资源 ★
+                try:
+                    if hasattr(cv_net, "resnet"):
+                        cv_net.resnet.to("cpu")
+                except Exception:
+                    pass
+                del cv_net
+                gc.collect()
+                if torch.cuda.is_available():
+                    torch.cuda.empty_cache()
+
+        return np.mean(fold_losses)
+
+    # ========= Optuna 调参 =========
+    def tune(self, max_evals: int = 50) -> None:
+        """
+        使用 Optuna 对 ResNet 做贝叶斯优化。
+        每个 trial 完成以后再做一次全局的显存清理。
+        """
+        def objective(trial: optuna.trial.Trial) -> float:
+            result = self.cross_val(trial)
+            # trial 级别兜底清理
+            gc.collect()
+            if torch.cuda.is_available():
+                torch.cuda.empty_cache()
+            return result
+
+        study = optuna.create_study(
+            direction='minimize',
+            sampler=optuna.samplers.TPESampler(seed=self.ctx.rand_seed)
+        )
+        study.optimize(objective, n_trials=max_evals)
+
+        self.best_params = study.best_params
+        self.best_trial = study.best_trial
+
+        params_path = self.output.result_path(
+            f'{self.ctx.model_nme}_bestparams_resn.csv'
+        )
+        pd.DataFrame(self.best_params, index=[0]).to_csv(params_path)
+
+    # ========= 用最优超参训练最终 ResNet =========
+    def train(self) -> None:
+        if not self.best_params:
+            raise RuntimeError('请先运行 tune() 以获得 ResNet 最优参数。')
+
+        self.model = ResNetSklearn(
+            model_nme=self.ctx.model_nme,
+            input_dim=self.ctx.train_oht_scl_data[self.ctx.var_nmes].shape[1],
+            task_type=self.ctx.task_type,
+            use_data_parallel=self.ctx.config.use_resn_data_parallel
+        )
+        self.model.set_params(self.best_params)
+
+        # 在全量 one-hot + 标准化数据上训练最终模型
+        self.model.fit(
+            self.ctx.train_oht_scl_data[self.ctx.var_nmes],
+            self.ctx.train_oht_scl_data[self.ctx.resp_nme],
+            self.ctx.train_oht_scl_data[self.ctx.weight_nme]
+        )
+
+        # 记录标签
+        self.ctx.model_label += [self.label]
+
+        # 训练集 / 测试集预测
+        self.ctx.train_data['pred_resn'] = self.model.predict(
+            self.ctx.train_oht_scl_data[self.ctx.var_nmes]
+        )
+        self.ctx.test_data['pred_resn'] = self.model.predict(
+            self.ctx.test_oht_scl_data[self.ctx.var_nmes]
+        )
+
+        # 加权赔付
+        self.ctx.train_data.loc[:, 'w_pred_resn'] = (
+            self.ctx.train_data['pred_resn'] *
+            self.ctx.train_data[self.ctx.weight_nme]
+        )
+        self.ctx.test_data.loc[:, 'w_pred_resn'] = (
+            self.ctx.test_data['pred_resn'] *
+            self.ctx.test_data[self.ctx.weight_nme]
+        )
+
+        # 方便外部调用
+        self.ctx.resn_best = self.model
+
+    # ========= 保存 / 加载 =========
+    def save(self) -> None:
+        """
+        只保存 ResNet 的 state_dict（轻量，不含优化器）。
+        """
+        if self.model is not None:
+            path = self.output.model_path(
+                f'01_{self.ctx.model_nme}_ResNet.pth'
+            )
+            torch.save(self.model.resnet.state_dict(), path)
+
+    def load(self) -> None:
+        """
+        从文件加载 ResNet 模型到合适的 device。
+        """
+        path = self.output.model_path(
+            f'01_{self.ctx.model_nme}_ResNet.pth'
+        )
+        if os.path.exists(path):
+            resn_loaded = ResNetSklearn(
+                model_nme=self.ctx.model_nme,
+                input_dim=self.ctx.train_oht_scl_data[self.ctx.var_nmes].shape[1],
+                task_type=self.ctx.task_type,
+                use_data_parallel=self.ctx.config.use_resn_data_parallel
+            )
+            state_dict = torch.load(path, map_location='cpu')
+            resn_loaded.resnet.load_state_dict(state_dict)
+
+            # 根据当前环境设置 device
+            if torch.cuda.is_available():
+                resn_loaded.device = torch.device('cuda')
+            elif torch.backends.mps.is_available():
+                resn_loaded.device = torch.device('mps')
+            else:
+                resn_loaded.device = torch.device('cpu')
+
+            resn_loaded.resnet.to(resn_loaded.device)
+            self.model = resn_loaded
+            self.ctx.resn_best = self.model
+        else:
+            print(f"[ResNetTrainer.load] 未找到模型文件：{path}")
+
+
+class FTTrainer(TrainerBase):
+    def __init__(self, context: "BayesOptModel") -> None:
+        if context.task_type == 'classification':
+            super().__init__(context, 'FTTransformerClassifier')
+        else:
+            super().__init__(context, 'FTTransformer')
+        self.model: Optional[FTTransformerSklearn] = None
+        self.best_params: Optional[Dict[str, Any]] = None
+        self.best_trial = None
+
+    def cross_val(self, trial: optuna.trial.Trial) -> float:
+        """
+        对 FT-Transformer 做交叉验证。
+        这里是显存最容易爆的地方，所以加入了：
+          - 较保守的超参搜索空间
+          - 每个 fold 结束后强制释放 GPU 显存
+        """
+        # 超参空间适当缩小一点，避免特别大的模型
+        learning_rate = trial.suggest_float(
+            'learning_rate', 1e-5, 5e-4, log=True
+        )
+        d_model = trial.suggest_int('d_model', 32, 256, step=32)
+        # n_heads = trial.suggest_categorical('n_heads', [2, 4]) 避免欠拟合
+        n_heads = trial.suggest_categorical('n_heads', [2, 4, 8])
+        # n_layers = trial.suggest_int('n_layers', 2, 4) 避免欠拟合
+        n_layers = trial.suggest_int('n_layers', 2, 8)
+        dropout = trial.suggest_float('dropout', 0.0, 0.2)
+        # batch_num = trial.suggest_int(
+        #     'batch_num',
+        #     5 if self.ctx.obj == 'reg:gamma' else 10,
+        #     10 if self.ctx.obj == 'reg:gamma' else 50,
+        #     step=1 if self.ctx.obj == 'reg:gamma' else 10
+        # )
+
+        if self.ctx.task_type == 'regression':
+            if self.ctx.obj == 'reg:tweedie':
+                tw_power = trial.suggest_float('tw_power', 1.0, 2.0)
+            elif self.ctx.obj == 'count:poisson':
+                tw_power = 1.0
+            elif self.ctx.obj == 'reg:gamma':
+                tw_power = 2.0
+            else:
+                tw_power = 1.5
+        else:  # classification
+            tw_power = None  # Not used
+
+        fold_losses = []
+
+        # 👉 可选：只在子样本上做 BO，避免大数据直接压垮显存
+        data_for_cv = self.ctx.train_data
+        max_rows_for_ft_bo = min(1000000, int(
+            len(data_for_cv)/2))   # 你可以根据显存情况调小或调大
+        if len(data_for_cv) > max_rows_for_ft_bo:
+            data_for_cv = data_for_cv.sample(
+                max_rows_for_ft_bo,
+                random_state=self.ctx.rand_seed
+            )
+
+        for _, (train_idx, test_idx) in enumerate(
+            self.ctx.cv.split(data_for_cv[self.ctx.factor_nmes])
+        ):
+            X_train_fold = data_for_cv.iloc[train_idx][self.ctx.factor_nmes]
+            y_train_fold = data_for_cv.iloc[train_idx][self.ctx.resp_nme]
+            w_train_fold = data_for_cv.iloc[train_idx][self.ctx.weight_nme]
+            X_val_fold = data_for_cv.iloc[test_idx][self.ctx.factor_nmes]
+            y_val_fold = data_for_cv.iloc[test_idx][self.ctx.resp_nme]
+            w_val_fold = data_for_cv.iloc[test_idx][self.ctx.weight_nme]
+
+            cv_ft = FTTransformerSklearn(
+                model_nme=self.ctx.model_nme,
+                num_cols=self.ctx.num_features,
+                cat_cols=self.ctx.cate_list,
+                d_model=d_model,
+                n_heads=n_heads,
+                n_layers=n_layers,
+                dropout=dropout,
+                task_type=self.ctx.task_type,
+                # batch_num=batch_num,
+                epochs=self.ctx.epochs,
+                tweedie_power=tw_power,
+                learning_rate=learning_rate,
+                patience=5,
+                use_data_parallel=self.ctx.config.use_ft_data_parallel
+            )
+
+            try:
+                cv_ft.fit(
+                    X_train_fold, y_train_fold, w_train_fold,
+                    X_val_fold, y_val_fold, w_val_fold
+                )
+                y_pred_fold = cv_ft.predict(X_val_fold)
+                if self.ctx.task_type == 'regression':
+                    loss = mean_tweedie_deviance(
+                        y_val_fold,
+                        y_pred_fold,
+                        sample_weight=w_val_fold,
+                        power=tw_power
+                    )
+                else:  # classification
+                    from sklearn.metrics import log_loss
+                    loss = log_loss(
+                        y_val_fold,
+                        y_pred_fold,
+                        sample_weight=w_val_fold,
+                    )
+                fold_losses.append(loss)
+            finally:
+                # 🧹 每个 fold 用完就立即释放 GPU 资源
+                try:
+                    # 如果模型在 GPU 上，先挪回 CPU
+                    if hasattr(cv_ft, "ft"):
+                        cv_ft.ft.to("cpu")
+                except Exception:
+                    pass
+                del cv_ft
+                gc.collect()
+                if torch.cuda.is_available():
+                    torch.cuda.empty_cache()
+
+        return np.mean(fold_losses)
+
+    def tune(self, max_evals: int = 50) -> None:
+        """
+        用 Optuna 做超参搜索。
+        在每个 trial 结束后再做一次显存清理，避免 trial 间显存碎片堆积。
+        """
+        def objective(trial: optuna.trial.Trial) -> float:
+            result = self.cross_val(trial)
+            # trial 级别的兜底清理
+            gc.collect()
+            if torch.cuda.is_available():
+                torch.cuda.empty_cache()
+            return result
+
+        study = optuna.create_study(
+            direction='minimize',
+            sampler=optuna.samplers.TPESampler(seed=self.ctx.rand_seed)
+        )
+        study.optimize(objective, n_trials=max_evals)
+        self.best_params = study.best_params
+        self.best_trial = study.best_trial
+        params_path = self.output.result_path(
+            f'{self.ctx.model_nme}_bestparams_ft.csv'
+        )
+        pd.DataFrame(self.best_params, index=[0]).to_csv(params_path)
+
+    def train(self) -> None:
+        if not self.best_params:
+            raise RuntimeError('请先运行 tune() 以获得 FT-Transformer 最优参数。')
+        self.model = FTTransformerSklearn(
+            model_nme=self.ctx.model_nme,
+            num_cols=self.ctx.num_features,
+            cat_cols=self.ctx.cate_list,
+            task_type=self.ctx.task_type,
+            use_data_parallel=self.ctx.config.use_ft_data_parallel
+        )
+        self.model.set_params(self.best_params)
+        self.model.fit(
+            self.ctx.train_data[self.ctx.factor_nmes],
+            self.ctx.train_data[self.ctx.resp_nme],
+            self.ctx.train_data[self.ctx.weight_nme]
+        )
+        self.ctx.model_label += [self.label]
+        self.ctx.train_data['pred_ft'] = self.model.predict(
+            self.ctx.train_data[self.ctx.factor_nmes]
+        )
+        self.ctx.test_data['pred_ft'] = self.model.predict(
+            self.ctx.test_data[self.ctx.factor_nmes]
+        )
+        self.ctx.train_data.loc[:, 'w_pred_ft'] = (
+            self.ctx.train_data['pred_ft'] *
+            self.ctx.train_data[self.ctx.weight_nme]
+        )
+        self.ctx.test_data.loc[:, 'w_pred_ft'] = (
+            self.ctx.test_data['pred_ft'] *
+            self.ctx.test_data[self.ctx.weight_nme]
+        )
+        self.ctx.ft_best = self.model
+
+    def save(self) -> None:
+        if self.model is not None:
+            torch.save(
+                self.model,
+                self.output.model_path(
+                    f'01_{self.ctx.model_nme}_FTTransformer.pth')
+            )
+
+    def load(self) -> None:
+        path = self.output.model_path(
+            f'01_{self.ctx.model_nme}_FTTransformer.pth')
+        if os.path.exists(path):
+            ft_loaded = torch.load(path, map_location='cpu')
+            if torch.cuda.is_available():
+                ft_loaded.device = torch.device('cuda')
+            elif torch.backends.mps.is_available():
+                ft_loaded.device = torch.device('mps')
+            else:
+                ft_loaded.device = torch.device('cpu')
+            ft_loaded.ft.to(ft_loaded.device)
+            self.model = ft_loaded
+            self.ctx.ft_best = self.model
+        else:
+            print(f"[load_model] Warning: 未找到 FT-Transformer 模型文件：{path}")
+
+
+# =============================================================================
+# BayesOpt orchestration & SHAP utilities
+# =============================================================================
+class BayesOptModel:
+    def __init__(self, train_data, test_data,
+                 model_nme, resp_nme, weight_nme, factor_nmes, task_type='regression',
+                 binary_resp_nme=None,
+                 cate_list=None, prop_test=0.25, rand_seed=None,
+                 epochs=100, use_gpu=True, 
+                 use_resn_data_parallel: bool = False, use_ft_data_parallel: bool = False):
+        cfg = BayesOptConfig(
+            model_nme=model_nme,
+            task_type=task_type,
+            resp_nme=resp_nme,
+            weight_nme=weight_nme,
+            factor_nmes=list(factor_nmes),
+            binary_resp_nme=binary_resp_nme,
+            cate_list=list(cate_list) if cate_list else None,
+            prop_test=prop_test,
+            rand_seed=rand_seed,
+            epochs=epochs,
+            use_gpu=use_gpu,
+            use_resn_data_parallel=use_resn_data_parallel,
+            use_ft_data_parallel=use_ft_data_parallel
+        )
+        self.config = cfg
+        self.model_nme = cfg.model_nme
+        self.task_type = cfg.task_type
+        self.resp_nme = cfg.resp_nme
+        self.weight_nme = cfg.weight_nme
+        self.factor_nmes = cfg.factor_nmes
+        self.binary_resp_nme = cfg.binary_resp_nme
+        self.cate_list = list(cfg.cate_list or [])
+        self.prop_test = cfg.prop_test
+        self.epochs = cfg.epochs
+        self.rand_seed = cfg.rand_seed if cfg.rand_seed is not None else np.random.randint(
+            1, 10000)
+        self.use_gpu = bool(cfg.use_gpu and torch.cuda.is_available())
+        self.output_manager = OutputManager(os.getcwd(), self.model_nme)
+
+        preprocessor = DatasetPreprocessor(train_data, test_data, cfg).run()
+        self.train_data = preprocessor.train_data
+        self.test_data = preprocessor.test_data
+        self.train_oht_scl_data = preprocessor.train_oht_scl_data
+        self.test_oht_scl_data = preprocessor.test_oht_scl_data
+        self.var_nmes = preprocessor.var_nmes
+        self.num_features = preprocessor.num_features
+        self.cat_categories_for_shap = preprocessor.cat_categories_for_shap
+
+        self.cv = ShuffleSplit(n_splits=int(1/self.prop_test),
+                               test_size=self.prop_test,
+                               random_state=self.rand_seed)
+        if self.task_type == 'classification':
+            self.obj = 'binary:logistic'
+        else:  # regression
+            if 'f' in self.model_nme:
+                self.obj = 'count:poisson'
+            elif 's' in self.model_nme:
+                self.obj = 'reg:gamma'
+            elif 'bc' in self.model_nme:
+                self.obj = 'reg:tweedie'
+            else:
+                self.obj = 'reg:tweedie'
+        self.fit_params = {
+            'sample_weight': self.train_data[self.weight_nme].values
+        }
+        self.model_label: List[str] = []
+
+        # 记录各模型训练器，后续统一通过标签访问，方便扩展新模型
+        self.trainers: Dict[str, TrainerBase] = {
+            'xgb': XGBTrainer(self),
+            'resn': ResNetTrainer(self),
+            'ft': FTTrainer(self),
+            'glm': GLMTrainer(self)
+        }
+        self.xgb_best = None
+        self.resn_best = None
+        self.glm_best = None
+        self.ft_best = None
+        self.best_xgb_params = None
+        self.best_resn_params = None
+        self.best_ft_params = None
+        self.best_xgb_trial = None
+        self.best_resn_trial = None
+        self.best_ft_trial = None
+        self.best_glm_params = None
+        self.best_glm_trial = None
+        self.xgb_load = None
+        self.resn_load = None
+        self.ft_load = None
+
+    # 定义单因素画图函数
+    def plot_oneway(self, n_bins=10):
+        for c in self.factor_nmes:
+            fig = plt.figure(figsize=(7, 5))
+            if c in self.cate_list:
+                group_col = c
+                plot_source = self.train_data
+            else:
+                group_col = f'{c}_bins'
+                bins = pd.qcut(
+                    self.train_data[c],
+                    n_bins,
+                    duplicates='drop'  # 注意：如果分位数重复会丢 bin，避免异常终止
+                )
+                plot_source = self.train_data.assign(**{group_col: bins})
+            plot_data = plot_source.groupby(
+                [group_col], observed=True).sum(numeric_only=True)
+            plot_data.reset_index(inplace=True)
+            plot_data['act_v'] = plot_data['w_act'] / \
+                plot_data[self.weight_nme]
+            plot_data.head()
+            ax = fig.add_subplot(111)
+            ax.plot(plot_data.index, plot_data['act_v'],
+                    label='Actual', color='red')
+            ax.set_title(
+                'Analysis of  %s : Train Data' % group_col,
+                fontsize=8)
+            plt.xticks(plot_data.index,
+                       list(plot_data[group_col].astype(str)),
+                       rotation=90)
+            if len(list(plot_data[group_col].astype(str))) > 50:
+                plt.xticks(fontsize=3)
+            else:
+                plt.xticks(fontsize=6)
+            plt.yticks(fontsize=6)
+            ax2 = ax.twinx()
+            ax2.bar(plot_data.index,
+                    plot_data[self.weight_nme],
+                    alpha=0.5, color='seagreen')
+            plt.yticks(fontsize=6)
+            plt.margins(0.05)
+            plt.subplots_adjust(wspace=0.3)
+            save_path = self.output_manager.plot_path(
+                f'00_{self.model_nme}_{group_col}_oneway.png')
+            plt.savefig(save_path, dpi=300)
+            plt.close(fig)
+
+    # 定义Xgboost贝叶斯优化函数
+    def bayesopt_xgb(self, max_evals=100):
+        trainer = self.trainers['xgb']
+        trainer.tune(max_evals)
+        trainer.train()
+        self.xgb_best = trainer.model
+        # 记录最优参数及 trial 以便排查或复现结果
+        self.best_xgb_params = trainer.best_params
+        self.best_xgb_trial = trainer.best_trial
+
+    # 定义GLM贝叶斯优化函数
+    def bayesopt_glm(self, max_evals=50):
+        trainer = self.trainers['glm']
+        trainer.tune(max_evals)
+        trainer.train()
+        self.glm_best = trainer.model
+        self.best_glm_params = trainer.best_params
+        self.best_glm_trial = trainer.best_trial
+
+    # 定义ResNet贝叶斯优化函数
+
+    def bayesopt_resnet(self, max_evals=100):
+        trainer = self.trainers['resn']
+        trainer.tune(max_evals)
+        trainer.train()
+        self.resn_best = trainer.model
+        # 保存最优 trial 相关信息，方便后续调参分析
+        self.best_resn_params = trainer.best_params
+        self.best_resn_trial = trainer.best_trial
+
+    # 定义 FT-Transformer 贝叶斯优化函数
+    def bayesopt_ft(self, max_evals=50):
+        trainer = self.trainers['ft']
+        trainer.tune(max_evals)
+        trainer.train()
+        self.ft_best = trainer.model
+        # FT-Transformer 参数较多，留存配置信息尤其重要
+        self.best_ft_params = trainer.best_params
+        self.best_ft_trial = trainer.best_trial
+
+    # 定义分箱函数
+
+    def _split_data(self, data, col_nme, wgt_nme, n_bins=10):
+        # 先按得分排序再按累计权重等分，能保证每个分箱曝光量接近
+        sorted_data = data.sort_values(by=col_nme, ascending=True).copy()
+        sorted_data['cum_weight'] = sorted_data[wgt_nme].cumsum()
+        w_sum = sorted_data[wgt_nme].sum()
+        if w_sum <= EPS:
+            sorted_data.loc[:, 'bins'] = 0
+        else:
+            sorted_data.loc[:, 'bins'] = np.floor(
+                sorted_data['cum_weight'] * float(n_bins) / w_sum
+            )
+            sorted_data.loc[(sorted_data['bins'] == n_bins),
+                            'bins'] = n_bins - 1
+        return sorted_data.groupby(['bins'], observed=True).sum(numeric_only=True)
+
+    # 构建提纯曲线所需的数据
+    def _plot_data_lift(self,
+                        pred_list, w_pred_list,
+                        w_act_list, weight_list, n_bins=10):
+        lift_data = pd.DataFrame()
+        lift_data.loc[:, 'pred'] = pred_list
+        lift_data.loc[:, 'w_pred'] = w_pred_list
+        lift_data.loc[:, 'act'] = w_act_list
+        lift_data.loc[:, 'weight'] = weight_list
+        plot_data = self._split_data(
+            lift_data, 'pred', 'weight', n_bins)
+        denom = np.maximum(plot_data['weight'], EPS)
+        plot_data['exp_v'] = plot_data['w_pred'] / denom
+        plot_data['act_v'] = plot_data['act'] / denom
+        plot_data.reset_index(inplace=True)
+        return plot_data
+
+    # 绘制提纯曲线
+    def plot_lift(self, model_label, pred_nme, n_bins=10):
+        # 绘制建模集上结果
+        figpos_list = [121, 122]
+        plot_dict = {
+            121: self.train_data,
+            122: self.test_data
+        }
+        name_list = {
+            121: 'Train Data',
+            122: 'Test Data'
+        }
+        if model_label == 'Xgboost':
+            pred_nme = 'pred_xgb'
+        elif model_label == 'ResNet':
+            pred_nme = 'pred_resn'
+        elif model_label == 'FTTransformer':
+            pred_nme = 'pred_ft'
+        elif model_label.startswith('ResNetClassifier'):
+            pred_nme = 'pred_resn'
+        elif model_label.startswith('FTTransformerClassifier'):
+            pred_nme = 'pred_ft'
+        elif model_label == 'GLM':
+            pred_nme = 'pred_glm'
+        # pred_nme 映射保证后续取列统一，否则新模型加入时需同步更新
+
+        fig = plt.figure(figsize=(11, 5))
+        for figpos in figpos_list:
+            plot_data = self._plot_data_lift(
+                plot_dict[figpos][pred_nme].values,
+                plot_dict[figpos]['w_'+pred_nme].values,
+                plot_dict[figpos]['w_act'].values,
+                plot_dict[figpos][self.weight_nme].values,
+                n_bins)
+            ax = fig.add_subplot(figpos)
+            ax.plot(plot_data.index, plot_data['act_v'],
+                    label='Actual', color='red')
+            ax.plot(plot_data.index, plot_data['exp_v'],
+                    label='Predicted', color='blue')
+            ax.set_title(
+                'Lift Chart on %s' % name_list[figpos], fontsize=8)
+            plt.xticks(plot_data.index,
+                       plot_data.index,
+                       rotation=90, fontsize=6)
+            plt.yticks(fontsize=6)
+            plt.legend(loc='upper left',
+                       fontsize=5, frameon=False)
+            plt.margins(0.05)
+            ax2 = ax.twinx()
+            ax2.bar(plot_data.index, plot_data['weight'],
+                    alpha=0.5, color='seagreen',
+                    label='Earned Exposure')
+            plt.yticks(fontsize=6)
+            plt.legend(loc='upper right',
+                       fontsize=5, frameon=False)
+            plt.subplots_adjust(wspace=0.3)
+            save_path = self.output_manager.plot_path(
+                f'01_{self.model_nme}_{model_label}_lift.png')
+            plt.savefig(save_path, dpi=300)
+        plt.show()
+        plt.close(fig)
+
+    # 构建双提纯曲线所需的数据
+    def _plot_data_dlift(self,
+                         pred_list_model1, pred_list_model2,
+                         w_list, w_act_list, n_bins=10):
+        lift_data = pd.DataFrame()
+        lift_data.loc[:, 'pred1'] = pred_list_model1
+        lift_data.loc[:, 'pred2'] = pred_list_model2
+        lift_data.loc[:, 'diff_ly'] = lift_data['pred1'] / lift_data['pred2']
+        lift_data.loc[:, 'act'] = w_act_list
+        lift_data.loc[:, 'weight'] = w_list
+        plot_data = self._split_data(lift_data, 'diff_ly', 'weight', n_bins)
+        denom = np.maximum(plot_data['act'], EPS)
+        plot_data['exp_v1'] = plot_data['pred1'] / denom
+        plot_data['exp_v2'] = plot_data['pred2'] / denom
+        plot_data['act_v'] = plot_data['act'] / denom
+        plot_data.reset_index(inplace=True)
+        return plot_data
+
+    # 绘制双提纯曲线
+    def plot_dlift(self, model_comp: List[str] = ['xgb', 'resn'], n_bins: int = 10) -> None:
+        """
+        绘制双提纯曲线，对比两个模型的预测效果。
+
+        Args:
+            model_comp: 包含两个模型简称的列表，例如 ['xgb', 'resn']。
+                        支持 'xgb', 'resn', 'ft'。
+            n_bins: 分箱数量。
+        """
+        if len(model_comp) != 2:
+            raise ValueError("`model_comp` 必须包含两个模型进行对比。")
+
+        model_name_map = {
+            'xgb': 'Xgboost',
+            'resn': 'ResNet',
+            'ft': 'FTTransformer',
+            'glm': 'GLM'
+        }
+
+        name1, name2 = model_comp
+        if name1 not in model_name_map or name2 not in model_name_map:
+            raise ValueError(f"不支持的模型简称。请从 {list(model_name_map.keys())} 中选择。")
+
+        fig, axes = plt.subplots(1, 2, figsize=(11, 5))
+        datasets = {
+            'Train Data': self.train_data,
+            'Test Data': self.test_data
+        }
+
+        for ax, (data_name, data) in zip(axes, datasets.items()):
+            pred1_col = f'w_pred_{name1}'
+            pred2_col = f'w_pred_{name2}'
+
+            if pred1_col not in data.columns or pred2_col not in data.columns:
+                print(
+                    f"警告: 在 {data_name} 中找不到预测列 {pred1_col} 或 {pred2_col}。跳过绘图。")
+                continue
+
+            plot_data = self._plot_data_dlift(
+                data[pred1_col].values,
+                data[pred2_col].values,
+                data[self.weight_nme].values,
+                data['w_act'].values,
+                n_bins
+            )
+
+            label1 = model_name_map[name1]
+            label2 = model_name_map[name2]
+
+            ax.plot(plot_data.index,
+                    plot_data['act_v'], label='Actual', color='red')
+            ax.plot(plot_data.index,
+                    plot_data['exp_v1'], label=label1, color='blue')
+            ax.plot(plot_data.index,
+                    plot_data['exp_v2'], label=label2, color='black')
+
+            ax.set_title(f'Double Lift Chart on {data_name}', fontsize=8)
+            ax.set_xticks(plot_data.index)
+            ax.set_xticklabels(plot_data.index, rotation=90, fontsize=6)
+            ax.set_xlabel(f'{label1} / {label2}', fontsize=6)
+            ax.tick_params(axis='y', labelsize=6)
+            ax.legend(loc='upper left', fontsize=5, frameon=False)
+            ax.margins(0.1)
+
+            ax2 = ax.twinx()
+            ax2.bar(plot_data.index, plot_data['weight'],
+                    alpha=0.5, color='seagreen', label='Earned Exposure')
+            ax2.tick_params(axis='y', labelsize=6)
+            ax2.legend(loc='upper right', fontsize=5, frameon=False)
+
+        plt.subplots_adjust(bottom=0.25, top=0.95, right=0.8, wspace=0.3)
+        save_path = self.output_manager.plot_path(
+            f'02_{self.model_nme}_dlift_{name1}_vs_{name2}.png')
+        plt.savefig(save_path, dpi=300)
+        plt.show()
+        plt.close(fig)
+
+    # 绘制成交率提升曲线
+    def plot_conversion_lift(self, model_pred_col: str, n_bins: int = 20):
+        if not self.binary_resp_nme:
+            print("错误: 未在 BayesOptModel 初始化时提供 `binary_resp_nme`。无法绘制成交率曲线。")
+            return
+
+        fig, axes = plt.subplots(1, 2, figsize=(14, 6), sharey=True)
+        datasets = {
+            'Train Data': self.train_data,
+            'Test Data': self.test_data
+        }
+
+        for ax, (data_name, data) in zip(axes, datasets.items()):
+            if model_pred_col not in data.columns:
+                print(f"警告: 在 {data_name} 中找不到预测列 '{model_pred_col}'。跳过绘图。")
+                continue
+
+            # 按模型预测分排序，并计算分箱
+            plot_data = data.sort_values(by=model_pred_col).copy()
+            plot_data['cum_weight'] = plot_data[self.weight_nme].cumsum()
+            total_weight = plot_data[self.weight_nme].sum()
+
+            if total_weight > EPS:
+                plot_data['bin'] = pd.cut(
+                    plot_data['cum_weight'],
+                    bins=n_bins,
+                    labels=False,
+                    right=False
+                )
+            else:
+                plot_data['bin'] = 0
+
+            # 按分箱聚合
+            lift_agg = plot_data.groupby('bin').agg(
+                total_weight=(self.weight_nme, 'sum'),
+                actual_conversions=(self.binary_resp_nme, 'sum'),
+                weighted_conversions=('w_binary_act', 'sum'),
+                avg_pred=(model_pred_col, 'mean')
+            ).reset_index()
+
+            # 计算成交率
+            lift_agg['conversion_rate'] = lift_agg['weighted_conversions'] / \
+                lift_agg['total_weight']
+
+            # 计算整体平均成交率
+            overall_conversion_rate = data['w_binary_act'].sum(
+            ) / data[self.weight_nme].sum()
+            ax.axhline(y=overall_conversion_rate, color='gray', linestyle='--',
+                       label=f'Overall Avg Rate ({overall_conversion_rate:.2%})')
+
+            ax.plot(lift_agg['bin'], lift_agg['conversion_rate'],
+                    marker='o', linestyle='-', label='Actual Conversion Rate')
+            ax.set_title(f'Conversion Rate Lift Chart on {data_name}')
+            ax.set_xlabel(f'Model Score Decile (based on {model_pred_col})')
+            ax.set_ylabel('Conversion Rate')
+            ax.grid(True, linestyle='--', alpha=0.6)
+            ax.legend()
+
+        plt.tight_layout()
+        plt.show()
+
+    # 保存模型
+
+    def save_model(self, model_name=None):
+
+        # model_name 可以是:
+        #   - None: 保存全部可用模型
+        #   - 'xgb': 只保存 Xgboost
+        #   - 'resn': 只保存 ResNet
+        #   - 'ft': 只保存 FT-Transformer
+        #   - 'glm': 只保存 GLM
+        if model_name in (None, 'xgb'):
+            trainer = self.trainers['xgb']
+            if trainer.model is not None:
+                trainer.save()
+            else:
+                print("[save_model] Warning: xgb_best 不存在,未保存 Xgboost 模型。")
+
+        if model_name in (None, 'resn'):
+            trainer = self.trainers['resn']
+            if trainer.model is not None:
+                trainer.save()
+            else:
+                print("[save_model] Warning: resn_best 不存在,未保存 ResNet 模型。")
+
+        if model_name in (None, 'ft'):
+            trainer = self.trainers['ft']
+            if trainer.model is not None:
+                trainer.save()
+            else:
+                print("[save_model] Warning: ft_best 不存在,未保存 FT-Transformer 模型。")
+
+        if model_name in (None, 'glm'):
+            trainer = self.trainers['glm']
+            if trainer.model is not None:
+                trainer.save()
+            else:
+                print("[save_model] Warning: glm_best 不存在,未保存 GLM 模型。")
+
+    def load_model(self, model_name=None):
+        # model_name 可以是:
+        #   - None: 加载全部能找到的模型
+        #   - 'xgb': 只加载 Xgboost
+        #   - 'resn': 只加载 ResNet
+        #   - 'ft': 只加载 FT-Transformer
+        #   - 'glm': 只加载 GLM
+
+        if model_name in (None, 'xgb'):
+            trainer = self.trainers['xgb']
+            trainer.load()
+            self.xgb_best = trainer.model
+            self.xgb_load = trainer.model
+
+        if model_name in (None, 'resn'):
+            trainer = self.trainers['resn']
+            trainer.load()
+            self.resn_best = trainer.model
+            self.resn_load = trainer.model
+
+        if model_name in (None, 'ft'):
+            trainer = self.trainers['ft']
+            trainer.load()
+            self.ft_best = trainer.model
+            self.ft_load = trainer.model
+
+        if model_name in (None, 'glm'):
+            trainer = self.trainers['glm']
+            trainer.load()
+            self.glm_best = trainer.model
+
+    def _build_ft_shap_matrix(self, data: pd.DataFrame) -> np.ndarray:
+
+        # 将原始特征 DataFrame (包含 self.factor_nmes) 转成
+        # 纯数值矩阵: 数值列为 float64，类别列为整数 code（float64 存储）。
+        # 列顺序与 self.factor_nmes 保持一致。
+
+        matrices = []
+
+        for col in self.factor_nmes:
+            s = data[col]
+
+            if col in self.cate_list:
+                # 类别列：按训练时的类别全集编码
+                cats = pd.Categorical(
+                    s,
+                    categories=self.cat_categories_for_shap[col]
+                )
+                # cats.codes 是一个 Index / ndarray，用 np.asarray 包一下再 reshape
+                codes = np.asarray(cats.codes, dtype=np.float64).reshape(-1, 1)
+                matrices.append(codes)
+            else:
+                # 数值列：转成 Series -> numpy -> reshape
+                vals = pd.to_numeric(s, errors="coerce")
+                arr = vals.to_numpy(dtype=np.float64, copy=True).reshape(-1, 1)
+                matrices.append(arr)
+
+        X_mat = np.concatenate(matrices, axis=1)  # (N, F)
+        return X_mat
+
+    def _decode_ft_shap_matrix_to_df(self, X_mat: np.ndarray) -> pd.DataFrame:
+
+        # 将 SHAP 的数值矩阵 (N, F) 还原为原始特征 DataFrame，
+        # 数值列为 float，类别列还原为 pandas 的 category 类型，
+        # 以便兼容 enable_categorical=True 的 XGBoost 和 FT-Transformer 的输入。
+        # 列顺序 = self.factor_nmes
+
+        data_dict = {}
+
+        for j, col in enumerate(self.factor_nmes):
+            col_vals = X_mat[:, j]
+
+            if col in self.cate_list:
+                cats = self.cat_categories_for_shap[col]
+
+                # SHAP 会扰动成小数，这里 round 回整数 code
+                codes = np.round(col_vals).astype(int)
+                # 限制在 [-1, len(cats)-1]
+                codes = np.clip(codes, -1, len(cats) - 1)
+
+                # 使用 pandas.Categorical.from_codes：
+                #   - codes = -1 被当成缺失 (NaN)
+                #   - 其他索引映射到 cats 中对应的类别
+                cat_series = pd.Categorical.from_codes(
+                    codes,
+                    categories=cats
+                )
+                # 存的是 Categorical 类型，而不是 object
+                data_dict[col] = cat_series
+            else:
+                # 数值列：直接 float
+                data_dict[col] = col_vals.astype(float)
+
+        df = pd.DataFrame(data_dict, columns=self.factor_nmes)
+
+        # 再保险：确保所有类别列 dtype 真的是 category
+        for col in self.cate_list:
+            if col in df.columns:
+                df[col] = df[col].astype("category")
+        return df
+
+    def _build_glm_design(self, data: pd.DataFrame) -> pd.DataFrame:
+        # 与 GLM 训练阶段一致：在 one-hot + 标准化特征上添加截距
+        X = data[self.var_nmes]
+        return sm.add_constant(X, has_constant='add')
+
+    # ========= XGBoost SHAP =========
+
+    def compute_shap_xgb(self, n_background: int = 500,
+                         n_samples: int = 200,
+                         on_train: bool = True):
+        # 使用 KernelExplainer 计算 XGBoost 的 SHAP 值（黑盒方式）。
+        #
+        # - 对 SHAP：输入是一份纯数值矩阵：
+        #     * 数值特征：float64
+        #     * 类别特征：用 _build_ft_shap_matrix 编码后的整数 code（float64）
+        # - 对模型：仍然用原始 DataFrame + xgb_best.predict(...)
+
+        if not hasattr(self, "xgb_best"):
+            raise RuntimeError("请先运行 bayesopt_xgb() 训练好 self.xgb_best")
+
+        # 1) 选择数据源：训练集 or 测试集（原始特征空间）
+        data = self.train_data if on_train else self.test_data
+        X_raw = data[self.factor_nmes]
+
+        # 2) 构造背景矩阵（用和 FT 一样的数值编码）
+        background_raw = X_raw.sample(
+            min(len(X_raw), n_background),
+            random_state=self.rand_seed
+        )
+        # KernelExplainer 计算量极大，务必控制背景样本规模，否则会拖慢调试
+        background_mat = self._build_ft_shap_matrix(
+            background_raw
+        ).astype(np.float64, copy=True)
+
+        # 3) 定义黑盒预测函数：数值矩阵 -> DataFrame -> xgb_best.predict
+        def f_predict(x_mat: np.ndarray) -> np.ndarray:
+            # 把编码矩阵还原成原始 DataFrame（数值+类别）
+            df_input = self._decode_ft_shap_matrix_to_df(x_mat)
+            # 注意：这里用的是 self.xgb_best.predict，和你训练/预测时一致
+            y_pred = self.xgb_best.predict(df_input)
+            return y_pred
+
+        explainer = shap.KernelExplainer(f_predict, background_mat)
+
+        # 4) 要解释的样本：原始特征 + 数值编码
+        X_explain_raw = X_raw.sample(
+            min(len(X_raw), n_samples),
+            random_state=self.rand_seed
+        )
+        X_explain_mat = self._build_ft_shap_matrix(
+            X_explain_raw
+        ).astype(np.float64, copy=True)
+
+        # 5) 计算 SHAP 值（注意用 nsamples='auto' 控制复杂度）
+        shap_values = explainer.shap_values(X_explain_mat, nsamples="auto")
+
+        # 6) 保存结果：
+        #   - shap_values：数值编码空间，对应 factor_nmes 的每一列
+        #   - X_explain_raw：原始 DataFrame，方便画图时显示真实类别名
+        self.shap_xgb = {
+            "explainer": explainer,
+            "X_explain": X_explain_raw,
+            "shap_values": shap_values,
+            "base_value": explainer.expected_value,
+        }
+        return self.shap_xgb
+    # ========= ResNet SHAP =========
+
+    def _resn_predict_wrapper(self, X_np):
+        # 保证走 CPU
+        model = self.resn_best.resnet.to("cpu")
+        with torch.no_grad():
+            # 不要 .to(self.device)
+            X_tensor = torch.tensor(X_np, dtype=torch.float32)
+            y_pred = model(X_tensor).cpu().numpy()
+        y_pred = np.clip(y_pred, 1e-6, None)
+        return y_pred.reshape(-1)
+
+    def compute_shap_resn(self, n_background: int = 500,
+                          n_samples: int = 200,
+                          on_train: bool = True):
+
+        # 使用 KernelExplainer 计算 ResNet 的 SHAP 值。
+        # 解释空间：已 one-hot & 标准化后的特征 self.var_nmes。
+
+        if not hasattr(self, 'resn_best'):
+            raise RuntimeError("请先运行 bayesopt_resnet() 训练好 resn_best")
+
+        self.resn_best.device = torch.device("cpu")  # 强制走 CPU
+        self.resn_best.resnet.to("cpu")
+        if torch.cuda.is_available():
+            torch.cuda.empty_cache()
+
+        # 选择数据集（已 one-hot & 标准化）
+        data = self.train_oht_scl_data if on_train else self.test_oht_scl_data
+        X = data[self.var_nmes]
+        if len(X) == 0:
+            raise ValueError(
+                "compute_shap_resn: 选择的数据集为空（len(X)==0），无法计算 SHAP。")
+
+        # 背景样本：float64 numpy
+        background_df = X.sample(
+            min(len(X), n_background),
+            random_state=self.rand_seed
+        )
+        background_np = background_df.to_numpy(dtype=np.float64, copy=True)
+
+        # 黑盒预测函数
+        def f_predict(x):
+            y = self._resn_predict_wrapper(x)
+            # 保证是一维数组
+            y = np.asarray(y, dtype=np.float64).reshape(-1)
+            return y
+
+        explainer = shap.KernelExplainer(f_predict, background_np)
+
+        # 要解释的样本
+        X_explain_df = X.sample(
+            min(len(X), n_samples),
+            random_state=self.rand_seed
+        )
+        X_explain_np = X_explain_df.to_numpy(dtype=np.float64, copy=True)
+
+        max_nsamples = 300
+        min_needed = X_explain_np.shape[1] + 2
+        nsample_eff = max(min_needed, min(max_nsamples,
+                                          X_explain_np.shape[0] * X_explain_np.shape[1]))
+        shap_values = explainer.shap_values(X_explain_np, nsamples=nsample_eff)
+        # 手动计算 base_value，避免 NotOneValueFound
+        bg_pred = f_predict(background_np)
+        if bg_pred.size == 0:
+            raise ValueError("compute_shap_resn: 背景样本预测结果为空，无法计算 base_value。")
+        base_value = float(bg_pred.mean())
+
+        self.shap_resn = {
+            "explainer": explainer,
+            "X_explain": X_explain_df,     # DataFrame: 用于画图（有列名）
+            "shap_values": shap_values,    # numpy: (n_samples, n_features)
+            # "base_value": explainer.expected_value,
+            "base_value": base_value,
+        }
+        return self.shap_resn
+
+    # ========= FT-Transformer SHAP =========
+
+    def _ft_shap_predict_wrapper(self, X_mat: np.ndarray) -> np.ndarray:
+
+        # SHAP 的预测包装：
+        # 数值矩阵 -> 还原为原始特征 DataFrame -> 调用 ft_best.predict
+
+        df_input = self._decode_ft_shap_matrix_to_df(X_mat)
+        y_pred = self.ft_best.predict(df_input)
+        return np.asarray(y_pred, dtype=np.float64).reshape(-1)
+
+    def compute_shap_ft(self, n_background: int = 500,
+                        n_samples: int = 200,
+                        on_train: bool = True):
+
+        # 使用 KernelExplainer 计算 FT-Transformer 的 SHAP 值。
+        # 解释空间：数值+类别code 的混合数值矩阵（float64），
+        # 但对外展示时仍使用原始特征名/取值（X_explain）。
+
+        if not hasattr(self, "ft_best"):
+            raise RuntimeError("请先运行 bayesopt_ft() 训练好 ft_best")
+
+        self.ft_best.device = torch.device("cpu")  # 强制走 CPU
+        self.ft_best.ft.to("cpu")
+        if torch.cuda.is_available():
+            torch.cuda.empty_cache()
+
+        # 选择数据源（原始特征空间）
+        data = self.train_data if on_train else self.test_data
+        X_raw = data[self.factor_nmes]
+
+        # 背景矩阵
+        background_raw = X_raw.sample(
+            min(len(X_raw), n_background),
+            random_state=self.rand_seed
+        )
+        background_mat = self._build_ft_shap_matrix(
+            background_raw
+        ).astype(np.float64, copy=True)
+
+        # 黑盒预测函数（数值矩阵 → DataFrame → FT 模型）
+        def f_predict(x):
+            return self._ft_shap_predict_wrapper(x)
+
+        explainer = shap.KernelExplainer(f_predict, background_mat)
+
+        # 要解释的样本（原始特征空间）
+        X_explain_raw = X_raw.sample(
+            min(len(X_raw), n_samples),
+            random_state=self.rand_seed
+        )
+        X_explain_mat = self._build_ft_shap_matrix(
+            X_explain_raw
+        ).astype(np.float64, copy=True)
+
+        max_nsamples = 300
+        min_needed = X_explain_mat.shape[1] + 2
+        nsample_eff = max(min_needed, min(max_nsamples,
+                                          X_explain_mat.shape[0] * X_explain_mat.shape[1]))
+        shap_values = explainer.shap_values(
+            X_explain_mat, nsamples=nsample_eff)
+        bg_pred = self._ft_shap_predict_wrapper(background_mat)
+        bg_pred = np.asarray(bg_pred, dtype=np.float64).reshape(-1)
+        base_value = float(bg_pred.mean())
+
+        self.shap_ft = {
+            "explainer": explainer,
+            "X_explain": X_explain_raw,   # 原始特征 DataFrame，用来画图
+            "shap_values": shap_values,   # numpy: (n_samples, n_features)
+            # "base_value": explainer.expected_value,
+            "base_value": base_value,
+        }
+        return self.shap_ft
+
+    # ========= GLM SHAP =========
+    def compute_shap_glm(self, n_background: int = 500,
+                         n_samples: int = 200,
+                         on_train: bool = True):
+        """
+        使用 KernelExplainer 计算 GLM 的 SHAP 值。
+        解释空间：one-hot + 标准化 + 截距项（与 GLM 训练一致）。
+        """
+        if not hasattr(self, "glm_best") or self.glm_best is None:
+            raise RuntimeError("请先运行 bayesopt_glm() 训练好 glm_best")
+
+        data = self.train_oht_scl_data if on_train else self.test_oht_scl_data
+        if len(data) == 0:
+            raise ValueError("compute_shap_glm: 选择的数据集为空，无法计算 SHAP。")
+
+        # 构造设计矩阵
+        design_all = self._build_glm_design(data)
+
+        # 背景样本
+        background_df = design_all.sample(
+            min(len(design_all), n_background),
+            random_state=self.rand_seed
+        )
+        background_np = background_df.to_numpy(dtype=np.float64, copy=True)
+        design_cols = list(design_all.columns)
+
+        def f_predict(x_np: np.ndarray) -> np.ndarray:
+            # 转成 DataFrame 以保持列顺序与截距
+            x_df = pd.DataFrame(x_np, columns=design_cols)
+            y_pred = self.glm_best.predict(x_df)
+            return np.asarray(y_pred, dtype=np.float64).reshape(-1)
+
+        explainer = shap.KernelExplainer(f_predict, background_np)
+
+        # 要解释的样本
+        explain_df = design_all.sample(
+            min(len(design_all), n_samples),
+            random_state=self.rand_seed
+        )
+        explain_np = explain_df.to_numpy(dtype=np.float64, copy=True)
+
+        max_nsamples = 300
+        min_needed = explain_np.shape[1] + 2
+        nsample_eff = max(min_needed, min(max_nsamples,
+                                          explain_np.shape[0] * explain_np.shape[1]))
+        shap_values = explainer.shap_values(explain_np, nsamples=nsample_eff)
+        bg_pred = f_predict(background_np)
+        base_value = float(np.asarray(bg_pred, dtype=np.float64).mean())
+
+        self.shap_glm = {
+            "explainer": explainer,
+            "X_explain": explain_df,      # 包含 const + 特征，用于画图
+            "shap_values": shap_values,
+            "base_value": base_value,
+            "design_columns": design_cols
+        }
+        return self.shap_glm