main_mmpd.py

import argparse
import os
import torch
import random
import numpy as np

from exp.exp_forecast import Exp_Forecast


from utils.tools import string_split, str2bool

import time

parser = argparse.ArgumentParser(description='MMPD Loss for Diverse Time Series Forecasting')

parser.add_argument('--data', type=str, required=True, default='ETTh1', help='data')
parser.add_argument('--root_path', type=str, default='./datasets/', help='root path of the data file')
parser.add_argument('--data_path', type=str, default='ETTh1.csv', help='data file')  
parser.add_argument('--data_split', type=str, default='0.7,0.1,0.2',help='train/val/test split, can be ratio or number')
parser.add_argument('--output_root', type=str, default='./out/', help='location to store output files')

#parameters for backbone
parser.add_argument('--backbone', type=str, default='Decoder', help='backbone model')
parser.add_argument('--in_len', type=int, default=336, help='input MTS length (T)')
parser.add_argument('--out_len', type=int, default=96, help='output MTS length (\tau)')
parser.add_argument('--patch_size', type=int, default=12, help='segment length (L_seg)')
parser.add_argument('--data_dim', type=int, default=7, help='Number of dimensions of the MTS data (D)')
parser.add_argument('--d_model', type=int, default=256, help='dimension of hidden states (d_model)')
parser.add_argument('--d_ff', type=int, default=512, help='dimension of MLP in transformer')
parser.add_argument('--n_heads', type=int, default=4, help='num of heads')
parser.add_argument('--e_layers', type=int, default=2, help='num of encoder layers (N)')
parser.add_argument('--d_layers', type=int, default=2, help='num of decoder layers (M)')
parser.add_argument('--dropout', type=float, default=0.2, help='dropout')

#parameters for loss function
parser.add_argument('--loss_func', type=str, default='MMPD', help='loss function')
parser.add_argument('--point_weight', type=float, default=0.01, help='weight for point loss')
parser.add_argument('--weighted', type=str2bool, default=True, help='weighted loss')
parser.add_argument('--d_diffusion', type=int, default=256, help='dimension for MLP in diffusion projector')
parser.add_argument('--diffusion_layers', type=int, default=1, help='num of diffusion layers')
parser.add_argument('--max_diffusion_steps', type=int, default=1000, help='max denosing steps')
parser.add_argument('--beta_schedule', type=str, default='linear', help='beta schedule for diffusion process')
parser.add_argument('--radius', type=int, default=3, help='radius for adjacent patches')

#parameters for training
parser.add_argument('--training', type=str2bool, default=True, help='training process')
parser.add_argument('--num_workers', type=int, default=4, help='data loader num workers')
parser.add_argument('--batch_size', type=int, default=32, help='batch size of train input data')
parser.add_argument('--train_epochs', type=int, default=20, help='train epochs')
parser.add_argument('--patience', type=int, default=5, help='early stopping patience')
parser.add_argument('--learning_rate', type=float, default=1e-4, help='optimizer initial learning rate')
parser.add_argument('--lradj', type=str, default='cosine',help='adjust learning rate')

#parameters for testing
parser.add_argument('--test_batch_num', type=int, default=-1, help='test batch number')

#for diffusion sampling
parser.add_argument('--testing', type=str2bool, default=True, help='testing process')
parser.add_argument('--prob_pred', type=str2bool, default=True, help='sample from diffusion')
parser.add_argument('--sample_num', type=int, default=100, help='sample number to compute expectation')
parser.add_argument('--num_sampling_steps', type=str, default='20', help='number of sampling steps for diffusion process')
parser.add_argument('--temperature', type=float, default=1.0, help='temperature for sampling')

#for diffusion gaussian mixture
parser.add_argument('--gmm_components', type=int, default=10, help='maximum number of components in GMM')
parser.add_argument('--prior_pi_decay', type=float, default=0.5, help='prior for weight decay, in the range of [0, 1], smaller value activates less components')
parser.add_argument('--prior_precision_shape', type=float, default=1e2, help='prior for variance')
parser.add_argument('--gmm_iterations', type=int, default=10, help='number of EM iterations for GMM at each diffusion step')

parser.add_argument('--use_gpu', type=str2bool, default=True, help='use gpu')
parser.add_argument('--gpu', type=int, default=0, help='gpu')
parser.add_argument('--use_multi_gpu', action='store_true', help='use multiple gpus', default=False)
parser.add_argument('--devices', type=str, default='0,1,2,3',help='device ids of multile gpus')

args = parser.parse_args()

args.use_gpu = True if torch.cuda.is_available() and args.use_gpu else False

if args.use_gpu and args.use_multi_gpu:
    args.devices = args.devices.replace(' ','')
    device_ids = args.devices.split(',')
    args.device_ids = [int(id_) for id_ in device_ids]
    args.gpu = args.device_ids[0]
    print(args.gpu)

data_parser = {
    'ETTh1':{'data':'ETTh1.csv', 'data_dim':7, 'split':[12*30*24, 4*30*24, 4*30*24]},
    'ETTm1':{'data':'ETTm1.csv', 'data_dim':7, 'split':[4*12*30*24, 4*4*30*24, 4*4*30*24]},
    'ETTh2':{'data':'ETTh2.csv', 'data_dim':7, 'split':[12*30*24, 4*30*24, 4*30*24]},
    'ETTm2':{'data':'ETTm2.csv', 'data_dim':7, 'split':[4*12*30*24, 4*4*30*24, 4*4*30*24]},
    'weather':{'data':'weather.csv', 'data_dim':21, 'split':[0.7, 0.1, 0.2]},
    'ECL':{'data':'electricity.csv', 'data_dim':321, 'split':[0.7, 0.1, 0.2]},
    'Traffic':{'data':'traffic.csv', 'data_dim':862, 'split':[0.7, 0.1, 0.2]},
    'dynamic':{'data':'dynamic_500K.csv', 'data_dim': 17, 'split':[0.7, 0.1, 0.2]},
}
if args.data in data_parser.keys():
    data_info = data_parser[args.data]
    args.data_path = data_info['data']
    args.data_dim = data_info['data_dim']
    args.data_split = data_info['split']
else:
    args.data_split = string_split(args.data_split)

print('Args in experiment:')
print(args)

#fix random seed
random_seed =2024
torch.manual_seed(random_seed)
random.seed(random_seed)
np.random.seed(random_seed)
torch.cuda.manual_seed(random_seed)
torch.cuda.manual_seed_all(random_seed)
torch.backends.cudnn.deterministic = True

Exp = Exp_Forecast

setting = 'data{}_il{}_ol{}_backbone{}_loss{}_weighted{}_patch{}_pointW{}_diffH{}_diffLayer{}_radius{}_diffStep{}_beta{}'.format(args.data, \
            args.in_len, args.out_len, args.backbone, args.loss_func, args.weighted,
            args.patch_size, args.point_weight, args.d_diffusion, \
            args.diffusion_layers, args.radius, args.max_diffusion_steps, args.beta_schedule)

exp = Exp(args) # set experiments
if args.training:
    print('>>>>>>>start training : {}>>>>>>>>>>>>>>>>>>>>>>>>>>'.format(setting))
    exp.train(setting)
if args.testing:
    print('>>>>>>>testing : {}<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<'.format(setting))
    exp.test(setting, test_batch_num=args.test_batch_num)