自回归滞后模型进行多变量时间序列预测

import pandas as pd
 
 # https://github.com/vcerqueira/blog/
 from src.tde import time_delay_embedding
 
 wine = pd.read_csv('data/wine_sales.csv', parse_dates=['date'])
 
 # setting date as index
 wine.set_index('date', inplace=True)
 
 # you can simulate some data with the following code
 # wine = pd.DataFrame(np.random.random((100, 6)),
 #           columns=['Fortified','Drywhite','Sweetwhite',
 #                     'Red','Rose','Sparkling'])
 
 # create data set with lagged features using time delay embedding
 wine_ds = []
 for col in wine:
  
  
 
 # concatenating all variables
 wine_df = pd.concat(wine_ds, axis=1).dropna()
 
 # defining target (Y) and explanatory variables (X)
 predictor_variables = wine_df.columns.str.contains('\(t\-')
 target_variables = wine_df.columns.str.contains('Sparkling\(t\+')
 
 X = wine_df.iloc[:, predictor_variables]
 Y = wine_df.iloc[:, target_variables]

from sklearn.model_selection import train_test_split
 from sklearn.metrics import mean_absolute_error as mae
 from sklearn.ensemble import RandomForestRegressor
 
 
 # train/test split
 X_tr, X_ts, Y_tr, Y_ts = train_test_split(X, Y, test_size=0.3, shuffle=False)
 
 # fitting a RF model
 model = RandomForestRegressor()
 model.fit(X_tr, Y_tr)
 
 # getting forecasts for the test set
 preds = model.predict(X_ts)
 
 # computing MAE error
 print(mae(Y_ts, preds))
 # 288.13

# getting importance scores from previous model
 importance_scores = pd.Series(dict(zip(X_tr.columns, model.feature_importances_)))
 
 # getting top 10 features
 top_10_features = importance_scores.sort_values(ascending=False)[:10]
 top_10_features_nm = top_10_features.index
 
 X_tr_top = X_tr[top_10_features_nm]
 X_ts_top = X_ts[top_10_features_nm]
 
 # re-fitting the model
 model_top_features = RandomForestRegressor()
 model_top_features.fit(X_tr_top, Y_tr)
 
 # getting forecasts for the test set
 preds_topf = model_top_features.predict(X_ts_top)
 
 # computing MAE error
 print(mae(Y_ts, preds_topf))
 # 274.36