-
Notifications
You must be signed in to change notification settings - Fork 0
/
train-stack.py
171 lines (141 loc) · 6.7 KB
/
train-stack.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
# 导入所需库
import joblib
import pandas as pd
import numpy as np
import xgboost as xgb
from sklearn.model_selection import train_test_split
from sklearn.preprocessing import LabelEncoder, StandardScaler
from sklearn.metrics import mean_squared_error, r2_score, mean_absolute_error
from sklearn.linear_model import LinearRegression
from sklearn.neighbors import KNeighborsRegressor
import torch
import torch.nn as nn
import torch.optim as optim
from hyperopt import fmin, tpe, hp, STATUS_OK, Trials
from tqdm import tqdm
# 固定 random_state 值
random_state_value = 42 # 选择任意整数值来保持一致性
# 自定义 MAPE 计算函数，避免除零问题
def mean_absolute_percentage_error(y_true, y_pred):
y_true, y_pred = np.array(y_true), np.array(y_pred)
non_zero_indices = y_true != 0
return np.mean(np.abs((y_true[non_zero_indices] - y_pred[non_zero_indices]) / y_true[non_zero_indices])) * 100
# 数据预处理
data = pd.read_excel('data2.xlsx')
data.replace({'#DIV/0!': 0}, inplace=True)
# 填充数值列中的 NaN 值
numeric_cols = data.select_dtypes(include=[np.number]).columns
data[numeric_cols] = data[numeric_cols].fillna(data[numeric_cols].mean())
label_encoder = LabelEncoder()
data['立地タイプ'] = label_encoder.fit_transform(data['立地タイプ'])
label_encoder_day_type = LabelEncoder()
data['曜日'] = label_encoder_day_type.fit_transform(data['曜日'])
scaler = StandardScaler()
numeric_cols = ['バスとの距離', '駅との距離', '人口_総数_300m以内', '男性割合', '15_64人口割合', '就業者_通学者割合',
'就業者_通学者利用交通手段_自転車割合']
data[numeric_cols] = scaler.fit_transform(data[numeric_cols])
joblib.dump(scaler, 'scaler.joblib')
joblib.dump(label_encoder, 'label_encoder.joblib')
joblib.dump(label_encoder_day_type, 'label_encoder_day_type.joblib')
X = data[['バスとの距離', '駅との距離', '立地タイプ', '曜日', '人口_総数_300m以内', '男性割合', '15_64人口割合',
'就業者_通学者割合', '就業者_通学者利用交通手段_自転車割合']]
y = data['利用回数']
# XGBoost 贝叶斯优化搜索空间
space = {
'n_estimators': hp.quniform('n_estimators', 100, 500, 50),
'max_depth': hp.quniform('max_depth', 3, 15, 1),
'learning_rate': hp.uniform('learning_rate', 0.01, 0.3),
'subsample': hp.uniform('subsample', 0.6, 1),
'colsample_bytree': hp.uniform('colsample_bytree', 0.6, 1),
'gamma': hp.uniform('gamma', 0, 0.5),
'min_child_weight': hp.quniform('min_child_weight', 1, 10, 1),
'reg_alpha': hp.uniform('reg_alpha', 0, 0.1),
'reg_lambda': hp.uniform('reg_lambda', 0.8, 1.2)
}
# 定义XGBoost目标函数
# 定义XGBoost目标函数
def objective(params):
params['max_depth'] = int(params['max_depth'])
params['n_estimators'] = int(params['n_estimators']) # 确保 n_estimators 是整数
model = xgb.XGBRegressor(objective='reg:squarederror', tree_method='hist', device='cuda', **params)
model.fit(X_train, y_train)
y_pred = model.predict(X_test)
mse = mean_squared_error(y_test, y_pred)
return {'loss': mse, 'status': STATUS_OK}
# RNN模型定义
class RNNModel(nn.Module):
def __init__(self, input_size, hidden_size, output_size):
super(RNNModel, self).__init__()
self.lstm = nn.LSTM(input_size, hidden_size, batch_first=True)
self.dropout = nn.Dropout(0.3)
self.fc = nn.Linear(hidden_size, output_size)
def forward(self, x):
lstm_out, _ = self.lstm(x)
out = self.dropout(lstm_out[:, -1, :])
return self.fc(out)
# 模型训练与堆叠
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
num_trials = 3
stacked_results = []
for i in range(num_trials):
print(f"第 {i + 1} 次模型训练")
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=random_state_value)
# XGBoost模型训练
trials = Trials()
best = fmin(fn=objective, space=space, algo=tpe.suggest, max_evals=200, trials=trials)
best_params = {
'n_estimators': int(best['n_estimators']), # 使用整数
'max_depth': int(best['max_depth']),
'learning_rate': best['learning_rate'],
'subsample': best['subsample'],
'colsample_bytree': best['colsample_bytree'],
'gamma': best['gamma']
}
xgb_model = xgb.XGBRegressor(objective='reg:squarederror', tree_method='hist', device='cuda', **best_params)
xgb_model.fit(X_train, y_train)
xgb_pred_train = xgb_model.predict(X_train)
xgb_pred_test = xgb_model.predict(X_test)
# KNN模型训练
knn_model = KNeighborsRegressor(n_neighbors=5)
knn_model.fit(X_train, y_train)
knn_pred_train = knn_model.predict(X_train)
knn_pred_test = knn_model.predict(X_test)
# RNN模型训练
X_train_rnn = torch.tensor(X_train.values, dtype=torch.float32).unsqueeze(2).to(device)
y_train_rnn = torch.tensor(y_train.values, dtype=torch.float32).unsqueeze(1).to(device)
X_test_rnn = torch.tensor(X_test.values, dtype=torch.float32).unsqueeze(2).to(device)
rnn_model = RNNModel(input_size=1, hidden_size=64, output_size=1).to(device)
criterion = nn.MSELoss()
optimizer = optim.Adam(rnn_model.parameters(), lr=0.001)
# 使用 tqdm 显示进度条
epochs = 100
for epoch in tqdm(range(epochs), desc="训练 RNN 模型"):
rnn_model.train()
optimizer.zero_grad()
outputs = rnn_model(X_train_rnn)
loss = criterion(outputs, y_train_rnn)
loss.backward()
optimizer.step()
rnn_model.eval()
with torch.no_grad():
rnn_pred_train = rnn_model(X_train_rnn).cpu().numpy().flatten()
rnn_pred_test = rnn_model(X_test_rnn).cpu().numpy().flatten()
# 合并模型预测结果
stacked_train = np.column_stack((xgb_pred_train, knn_pred_train, rnn_pred_train))
stacked_test = np.column_stack((xgb_pred_test, knn_pred_test, rnn_pred_test))
# 使用线性回归作为元学习器
stacked_model = LinearRegression()
stacked_model.fit(stacked_train, y_train)
final_predictions = stacked_model.predict(stacked_test)
# 计算评估指标
mse = mean_squared_error(y_test, final_predictions)
rmse = np.sqrt(mse)
mae = mean_absolute_error(y_test, final_predictions)
mape = mean_absolute_percentage_error(y_test, final_predictions)
r2 = r2_score(y_test, final_predictions)
print(f"第 {i + 1} 次集成模型的 MSE: {mse}")
print(f"第 {i + 1} 次集成模型的 RMSE: {rmse}")
print(f"第 {i + 1} 次集成模型的 MAE: {mae}")
print(f"第 {i + 1} 次集成模型的 MAPE: {mape:.2f}%")
print(f"第 {i + 1} 次集成模型的 R²: {r2}")
stacked_results.append((final_predictions, mse, rmse, mae, mape, r2))