test: 对比时间顺序划分特征方案

outputs/time_split_experiment/time_split_feature_model_comparison.csv

@@ -0,0 +1,17 @@
+特征方案,模型,特征数,训练集样本数,测试集样本数,训练日期起,训练日期止,测试日期起,测试日期止,RMSE,MAE,R2,测试预测最小值,测试预测最大值,未来预测最小值,未来预测最大值,未来预测均值
+13特征_日期sin_cos,梯度提升树,13,584,147,2011-01-01,2012-08-06,2012-08-07,2012-12-31,1035.393806499406,869.0366952335274,0.6949404951600915,1193.005865875163,7448.828701180003,1862,6350,3848.46
+12特征_仅日期sin,梯度提升树,12,584,147,2011-01-01,2012-08-06,2012-08-07,2012-12-31,1037.9787565353552,864.907988042832,0.6934153789548839,1570.2284280500735,7557.904184096963,1752,6174,3687.28
+11特征_无日期,梯度提升树,11,584,147,2011-01-01,2012-08-06,2012-08-07,2012-12-31,1065.8819685946905,903.8377254645945,0.6767104504461063,1526.0106661213501,7445.060381049142,1927,6896,4029.71
+12特征_日期数值,线性回归,12,584,147,2011-01-01,2012-08-06,2012-08-07,2012-12-31,1100.6960028141114,802.9604968933488,0.6552468729685021,1667.5505098104527,7202.426019549715,1341,5571,3322.7
+11特征_无日期,线性回归,11,584,147,2011-01-01,2012-08-06,2012-08-07,2012-12-31,1102.6619968627012,803.6253768758826,0.6540142203549053,1677.5487160817047,7191.784877243726,1775,6001,3756.0
+12特征_仅日期sin,线性回归,12,584,147,2011-01-01,2012-08-06,2012-08-07,2012-12-31,1109.0855745651786,812.0301896212464,0.6499713857316858,1680.4361958712066,7184.255023908034,1772,6010,3754.92
+13特征_日期sin_cos,线性回归,13,584,147,2011-01-01,2012-08-06,2012-08-07,2012-12-31,1111.9746531184803,816.5536163460256,0.6481454179049195,1691.5799618779006,7248.332307970662,1775,5909,3690.63
+12特征_日期数值,梯度提升树,12,584,147,2011-01-01,2012-08-06,2012-08-07,2012-12-31,1115.505147894367,754.6104141829828,0.6459076101057781,2210.8052037892908,7886.894463218566,3884,8023,6177.22
+12特征_日期数值,随机森林,12,584,147,2011-01-01,2012-08-06,2012-08-07,2012-12-31,1211.6404640631436,857.2259950393145,0.5822456578217434,3362.6023664502195,7437.1805507936515,3778,7634,5700.92
+11特征_无日期,随机森林,11,584,147,2011-01-01,2012-08-06,2012-08-07,2012-12-31,1243.2258138091481,1038.575310261998,0.5601815197920028,2508.394888672439,7396.9488500000025,1991,6107,3698.12
+11特征_无日期,SVR,11,584,147,2011-01-01,2012-08-06,2012-08-07,2012-12-31,1256.1142258603181,1110.5200812033272,0.5510151326775178,2435.653600556283,6728.154381038747,1857,5138,3785.22
+12特征_仅日期sin,随机森林,12,584,147,2011-01-01,2012-08-06,2012-08-07,2012-12-31,1300.1505289996633,1061.3027606571065,0.5189826853858082,2349.546295021645,7412.202957142856,1966,5800,3536.75
+12特征_日期数值,SVR,12,584,147,2011-01-01,2012-08-06,2012-08-07,2012-12-31,1302.8065606379432,1122.631254189851,0.517015371360585,3038.4325167245174,6793.0357932052375,3203,5408,4553.72
+13特征_日期sin_cos,随机森林,13,584,147,2011-01-01,2012-08-06,2012-08-07,2012-12-31,1362.4036579408184,1162.7424853677699,0.47181624787772836,2337.0508269841275,7378.615926190471,1981,5446,3498.99
+13特征_日期sin_cos,SVR,13,584,147,2011-01-01,2012-08-06,2012-08-07,2012-12-31,1500.6745922933299,1355.960815132986,0.35916459954458013,2458.0732583146823,6405.637428945096,1864,4975,3596.58
+12特征_仅日期sin,SVR,12,584,147,2011-01-01,2012-08-06,2012-08-07,2012-12-31,1514.2320642644604,1364.971543352133,0.3475333594320331,2414.919176938589,6342.6982615190545,1813,4998,3673.18

time_split_experiment.py

@@ -0,0 +1,140 @@
+from pathlib import Path
+
+import numpy as np
+import pandas as pd
+from sklearn.compose import ColumnTransformer
+from sklearn.ensemble import GradientBoostingRegressor, RandomForestRegressor
+from sklearn.linear_model import LinearRegression
+from sklearn.metrics import mean_absolute_error, mean_squared_error, r2_score
+from sklearn.pipeline import Pipeline
+from sklearn.preprocessing import OneHotEncoder, StandardScaler
+from sklearn.svm import SVR
+
+BASE_DIR = Path(__file__).resolve().parent
+DATA_PATH = Path('/root/.hermes-web-ui/upload/default/28752a235eb6ef9d.xlsx')
+OUT_DIR = BASE_DIR / 'outputs' / 'time_split_experiment'
+OUT_DIR.mkdir(parents=True, exist_ok=True)
+
+TARGET = '总骑行量'
+BASE_FEATURES = ['年', '季节', '月', '节假日', '星期', '工作日', '天气', '气温', '体感温度', '湿度', '风速']
+CATEGORICAL_BASE = ['年', '季节', '月', '节假日', '星期', '工作日', '天气']
+
+
+def rmse(y_true, y_pred):
+    return float(np.sqrt(mean_squared_error(y_true, y_pred)))
+
+
+def make_pipeline(name, estimator, features, categorical):
+    numeric = [col for col in features if col not in categorical]
+    need_scale = name in {'线性回归', 'SVR'}
+    preprocessor = ColumnTransformer(
+        transformers=[
+            ('cat', OneHotEncoder(handle_unknown='ignore'), categorical),
+            ('num', StandardScaler() if need_scale else 'passthrough', numeric),
+        ],
+        verbose_feature_names_out=False,
+    )
+    return Pipeline([('preprocess', preprocessor), ('model', estimator)])
+
+
+def add_date_features(frame, min_date):
+    frame = frame.copy()
+    frame['日期数值'] = (frame['日期'] - min_date).dt.days
+    day_of_year = frame['日期'].dt.dayofyear
+    frame['日期_sin'] = np.sin(2 * np.pi * day_of_year / 365.25)
+    frame['日期_cos'] = np.cos(2 * np.pi * day_of_year / 365.25)
+    return frame
+
+
+def main():
+    daily = pd.read_excel(DATA_PATH, sheet_name='每天数据')
+    future = pd.read_excel(DATA_PATH, sheet_name='预测集')
+    daily['日期'] = pd.to_datetime(daily['日期'])
+    future['日期'] = pd.to_datetime(future['日期'])
+    daily = daily.sort_values('日期').reset_index(drop=True)
+    future = future.sort_values('日期').reset_index(drop=True)
+
+    min_date = daily['日期'].min()
+    daily = add_date_features(daily, min_date)
+    future = add_date_features(future, min_date)
+
+    split_idx = int(len(daily) * 0.8)
+    train = daily.iloc[:split_idx].copy()
+    test = daily.iloc[split_idx:].copy()
+
+    feature_sets = {
+        '11特征_无日期': {
+            'features': BASE_FEATURES,
+            'categorical': CATEGORICAL_BASE,
+        },
+        '12特征_日期数值': {
+            'features': ['日期数值'] + BASE_FEATURES,
+            'categorical': CATEGORICAL_BASE,
+        },
+        '13特征_日期sin_cos': {
+            'features': ['日期_sin', '日期_cos'] + BASE_FEATURES,
+            'categorical': CATEGORICAL_BASE,
+        },
+        '12特征_仅日期sin': {
+            'features': ['日期_sin'] + BASE_FEATURES,
+            'categorical': CATEGORICAL_BASE,
+        },
+    }
+
+    model_factories = {
+        '线性回归': lambda: LinearRegression(),
+        '随机森林': lambda: RandomForestRegressor(n_estimators=500, random_state=42, min_samples_leaf=2, n_jobs=-1),
+        '梯度提升树': lambda: GradientBoostingRegressor(random_state=42, n_estimators=300, learning_rate=0.05, max_depth=3),
+        'SVR': lambda: SVR(C=200, gamma='scale', epsilon=50),
+    }
+
+    rows = []
+    future_rows = []
+    for fs_name, spec in feature_sets.items():
+        features = spec['features']
+        categorical = spec['categorical']
+        for model_name, factory in model_factories.items():
+            pipe = make_pipeline(model_name, factory(), features, categorical)
+            pipe.fit(train[features], train[TARGET])
+            pred = pipe.predict(test[features])
+            future_pred = np.maximum(np.round(pipe.predict(future[features])), 0).astype(int)
+            rows.append({
+                '特征方案': fs_name,
+                '模型': model_name,
+                '特征数': len(features),
+                '训练集样本数': len(train),
+                '测试集样本数': len(test),
+                '训练日期起': train['日期'].min().date().isoformat(),
+                '训练日期止': train['日期'].max().date().isoformat(),
+                '测试日期起': test['日期'].min().date().isoformat(),
+                '测试日期止': test['日期'].max().date().isoformat(),
+                'RMSE': rmse(test[TARGET], pred),
+                'MAE': float(mean_absolute_error(test[TARGET], pred)),
+                'R2': float(r2_score(test[TARGET], pred)),
+                '测试预测最小值': float(pred.min()),
+                '测试预测最大值': float(pred.max()),
+                '未来预测最小值': int(future_pred.min()),
+                '未来预测最大值': int(future_pred.max()),
+                '未来预测均值': float(future_pred.mean()),
+            })
+            if model_name == '梯度提升树':
+                temp = future[['日期'] + BASE_FEATURES].copy()
+                temp['特征方案'] = fs_name
+                temp['模型'] = model_name
+                temp['预测总骑行量'] = future_pred
+                future_rows.append(temp)
+
+    result = pd.DataFrame(rows).sort_values(['RMSE', 'R2'], ascending=[True, False])
+    result.to_csv(OUT_DIR / 'time_split_feature_model_comparison.csv', index=False, encoding='utf-8-sig')
+    pd.concat(future_rows, ignore_index=True).to_excel(OUT_DIR / 'gbdt_future_predictions_by_feature_set.xlsx', index=False)
+
+    print('时间顺序划分：')
+    print(f"训练集 {len(train)} 条：{train['日期'].min().date()} 至 {train['日期'].max().date()}")
+    print(f"测试集 {len(test)} 条：{test['日期'].min().date()} 至 {test['日期'].max().date()}")
+    print('\n按 RMSE 排序：')
+    print(result[['特征方案', '模型', 'RMSE', 'MAE', 'R2', '未来预测最小值', '未来预测最大值', '未来预测均值']].to_string(index=False, float_format=lambda x: f'{x:.3f}'))
+    print(f"\n结果已保存：{OUT_DIR}")
+
+
+if __name__ == '__main__':
+    main()