LSTM时间序列预测 LSTM时间序列预测

'''导包''' import pandas as pd import numpy as np import tqdm import datetime import os import argparse import random import creat_data as Data from def_function import * import matplotlib.pyplot as plt import torch import torch.nn as nn import torch.utils.data import logging import sys from config import Config ''' 主函数 args: 超参定义器,param logger: 日志句柄''' def main(args, logger): device = torch.device(args.device) # 指定运行设备 data_x, data_y, scalar = get_dataset(args.data_path, args.dataset_name, args.city_name, args.look_back) ''' 划分训练集和测试集的方法，但这么做会影响时序，最好还是利用新的测试集 ''' train_slice = int(data_x.shape[0] * 0.8) # 8比2划分训练集 切片 前80%的data_x定义为train_slice train_dataset = Data.generate_dataset(data_x[:train_slice, :], data_y[:train_slice, :]) # 建立训练集数据库 train_data_loader = torch.utils.data.DataLoader(train_dataset, batch_size=args.batch_size, num_workers=args.num_workers,shuffle=False) # 批次训练 test_dataset = Data.generate_dataset(data_x[train_slice:, :], data_y[train_slice:, :]) # 建立测试集数据库 test_data_loader = torch.utils.data.DataLoader(test_dataset, batch_size=args.batch_size, num_workers=args.num_workers) # 批次测试 #train_dataset = Data.generate_dataset(data_x, data_y) # 以整个导入的数据全部作为训练集 #train_data_loader = torch.utils.data.DataLoader(train_dataset, batch_size=args.batch_size, num_workers=args.num_workers,shuffle=False) #不打乱数据 批次训练 feature_nums = args.look_back model = get_model(args.model_name, feature_nums, args.hidden_dims, args.num_layers).to(device) logger.info(model) # 日志 if args.loss_type == 'mse': #损失函数类型 mse和smoothl1loss loss = nn.MSELoss() elif args.loss_type == 'smoothl1loss': loss = nn.SmoothL1Loss() optimizer = torch.optim.Adam(params=model.parameters(), lr=args.lr, weight_decay=args.weight_decay) # 优化器采用Adam start_time = datetime.datetime.now() # 起始时间 train_epoch_loss = [] # 用数组存放每epoch的损失 for epoch_i in range(args.epochs): torch.cuda.empty_cache() # 清理无用的cuda中间变量缓存 train_average_loss = train(model, optimizer, train_data_loader, loss, device) # 平均损失 train_epoch_loss.append(train_average_loss) train_end_time = datetime.datetime.now() # 训练时间 if epoch_i % args.print_interval == 0: # 每print_interval，打印一次结果到日志中 logger.info('City {}, model {}, epoch {}, train_{}_loss {}, ' '[{}s]'.format(args.city_name, args.model_name, epoch_i, args.loss_type, train_average_loss, (train_end_time - start_time).seconds)) torch.save(model.state_dict(), os.path.join(args.save_param_dir, args.model_name + '_best_' + args.loss_type + '.pth')) # 存储参数 '''valid_dataset = Data.generate_dataset(data_x,data_y) # 验证集数据 valid_data_loader = torch.utils.data.DataLoader(valid_dataset, batch_size=args.batch_size, num_workers=args.num_workers)''' # submission #valid_loss, preds = sub(model, valid_data_loader, loss, device) ''' 如果从训练集中划分了测试集，这里可以改为如下形式，test_data_loader为上面第146行定义的变量 ''' valid_loss, preds = sub(model, test_data_loader, loss, device) output_dict = {} output_dict.setdefault(args.loss_type + '_loss', train_epoch_loss) output_dict.setdefault(args.loss_type + '_preds', (preds * scalar).tolist()) return output_dict if __name__ == '__main__': #模拟程序入口 使用的是多线程导入数据，用 __name__这样写不会报错 parser = argparse.ArgumentParser() #创建解析器 解析config.py parser.add_argument('--data_path', default=Config.data_path) #也可使用绝对路径，默认是config中的相对路径 parser.add_argument('--dataset_name', default='yjz.csv', help='dataset') parser.add_argument('--model_name', default=Config.model_name, help='LR, RNN, MLP') parser.add_argument('--num_workers', default=Config.num_workers, help='4, 8, 16, 32') parser.add_argument('--hidden_dims', default=Config.hidden_dims) parser.add_argument('--num_layers', default=Config.num_layers, help='1, 2') parser.add_argument('--look_back', default=Config.look_back, help='以几行数据为特征维度数量') parser.add_argument('--city_name', default='yjz5', help='Zone 1,Zone 2,Zone 3') parser.add_argument('--epochs', type=int, default=Config.epochs) parser.add_argument('--lr', type=float, default=Config.lr) parser.add_argument('--weight_decay', type=float, default=Config.weight_decay) parser.add_argument('--batch_size', type=int, default=Config.batch_size) parser.add_argument('--print_interval', type=int, default=Config.print_interval) parser.add_argument('--device', default=Config.device) parser.add_argument('--loss_type', type=str, default='mse', help='smoothl1loss') parser.add_argument('--save_log_dir', default=Config.save_log_dir) parser.add_argument('--save_res_dir', default=Config.save_res_dir) parser.add_argument('--save_param_dir', default=Config.save_param_dir) args = parser.parse_args() # 设置随机数种子 setup_seed(Config.seed) #创建保存结果文件夹 if not os.path.exists(args.save_log_dir): os.mkdir(args.save_log_dir) if not os.path.exists(args.save_res_dir): os.mkdir(args.save_res_dir) if not os.path.exists(args.save_param_dir): os.mkdir(args.save_param_dir) logging.basicConfig(level=logging.DEBUG, filename=args.save_log_dir + args.model_name + '_output.log', datefmt='%Y/%m/%d %H:%M:%S', format='%(asctime)s - %(name)s - %(levelname)s - %(lineno)d - %(module)s - %(message)s') logger = logging.getLogger(__name__) logger.setLevel(logging.DEBUG) stream_handler = logging.StreamHandler(sys.stdout) stream_handler.setLevel(logging.INFO) logger.addHandler(stream_handler) logger.info('===> start training! ') # 将各个州的数据保存到字典中 current_city_output_dicts = dict.fromkeys(tuple('Zone 2,Zone 3'.split(','))) #Zone2 3 也可加上 'Zone 1,Zone 2,Zone 3' for city_name in current_city_output_dicts.keys(): logger.info('===> now excuate the city {} '.format(city_name)) args.city_name = city_name current_city_output_dict = {} for loss_type in ['mse', 'smoothl1loss']: args.loss_type = loss_type output_dict = main(args, logger) # 主运行函数 for key in output_dict.keys(): current_city_output_dict.setdefault(key, output_dict[key]) current_city_output_dicts[city_name] = current_city_output_dict # 存储loss和preds for loss_type in ['mse', 'smoothl1loss']: city_preds = {} city_losses = {} for city_name in current_city_output_dicts.keys(): current_city_output_loss = current_city_output_dicts[city_name][loss_type + '_loss'] current_city_output_pred = current_city_output_dicts[city_name][loss_type + '_preds'] city_losses.setdefault(city_name, current_city_output_loss) city_preds.setdefault(city_name, current_city_output_pred) city_loss_df = pd.DataFrame(data=city_losses) city_loss_df.to_csv(args.save_res_dir + '/ele_' + loss_type + '_losses_' + args.model_name + '.csv', index=None)