PatchTST 벤치마크 결과

본 포스팅은 PatchTST 벤치마크 결과에 대한 내용입니다.

---

 

PatchTST 벤치마크 코드 (Google Colab)

저번 PatchTST의 이론적인 설명을 기반으로하여 모델을 구현 및 학습을 통해 성능 결과와 시각화를 출력해보겠습니다.  

!git clone https://github.com/yuqinie98/PatchTST.git
!git clone https://github.com/zhouhaoyi/ETDataset.git
!ls

import sys
if not 'PatchTST' in sys.path:
sys.path += ['PatchTST']
sys.path

from utils.tools import dotdict
from exp.exp_main import Exp_Main
import torch

args = dotdict()

# 모델과 데이터셋 설정
args.model_id = 'test'
args.model_name = 'PatchTST'
args.data = 'ETTh1'
args.root_path = '/content/ETDataset/ETT-small/'
args.data_path = 'ETTh1.csv'
args.features = 'M'
args.target = 'OT'
args.freq = 'h'
args.checkpoints = '/content/checkpoints'
args.is_training = 1
# 학습 설정
args.random_seed = 2021
args.seq_len = 512
args.label_len = 0
args.pred_len = 96

# former 모델 구조 설정
args.model = 'PatchTST'
args.embed_type = 0
args.enc_in = 7
args.dec_in = 7
args.c_out = 7
args.n_heads = 4
args.d_model = 128
args.e_layers = 10
args.d_layers= 2
args.d_ff = 128
args.moving_avg = 100
args.factor = 1
args.distill = True
args.dropout = 0.3
args.embed = 'timeF'
args.activation = 'gelu'
args.output_attention = True
args.do_predict = True

# PatchTST 모델 구조 설정
args.fc_dropout = 0.3
args.head_dropout = 0.5
args.patch_len = 16
args.stride = 8
args.padding_patch = 'end'
args.revin = 1
args.affine = 0
args.subtract_last = 0
args.decomposition = 1
args.kernel_size = 25
args.individual = 0
args.store_attn=True

# 학습 과정 설정
args.train_epochs = 100
args.itr = 1
args.batch_size = 128
args.learning_rate = 0.0001
args.num_workers = 30
args.pct_start = 0.3
args.use_amp = False
args.des = 'Exp'
args.loss = 'mse'
args.lradj = 'type3'
args.patience = 30

# GPU
args.use_gpu = True
args.gpu = 0
args.devices = '0'
args.use_multi_gpu = False

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]

args.detail_freq = args.freq
args.freq = args.freq[-1:]

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

Exp = Exp_Main

if args.is_training:
for ii in range(args.itr):
# setting record of experiments
setting = '{}_{}_{}_ft{}_sl{}_ll{}_pl{}_dm{}_nh{}_el{}_dl{}_df{}_fc{}_eb{}_dt{}_{}_{}'.format(
args.model_id,
args.model,
args.data,
args.features,
args.seq_len,
args.label_len,
args.pred_len,
args.d_model,
args.n_heads,
args.e_layers,
args.d_layers,
args.d_ff,
args.factor,
args.embed,
args.distil,
args.des,ii,)

  exp = Exp(args)  # set experiments
  print('>>>>>>>start training : {}>>>>>>>>>>>>>>>>>>>>>>>>>>'.format(setting))
  exp.train(setting)

  print('>>>>>>>testing : {}<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<'.format(setting))
  exp.test(setting)

  if args.do_predict:
      print('>>>>>>>predicting : {}<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<'.format(setting))
      exp.predict(setting, True)

  torch.cuda.empty_cache()
  else:
  ii = 0
  setting = '{}_{}_{}_ft{}_sl{}_ll{}_pl{}_dm{}_nh{}_el{}_dl{}_df{}_fc{}_eb{}_dt{}_{}_{}'.format(args.model_id,
  args.model,
  args.data,
  args.features,
  args.seq_len,
  args.label_len,
  args.pred_len,
  args.d_model,
  args.n_heads,
  args.e_layers,
  args.d_layers,
  args.d_ff,
  args.factor,
  args.embed,
  args.distil,
  args.des, ii)
  

exp = Exp(args)  # set experiments
exp = Exp(args)  # set experiments
print('>>>>>>>testing : {}<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<'.format(setting))
exp.test(setting, test=1)
torch.cuda.empty_cache()

모델 테스트 성능 결과

 

모델의 성능 지표는 MAE와 MSE로 측정되며 다음과 같이 성능을 증명하였습니다. mse:0.4166485071182251, mae:0.43856024742126465

• 해당 모델의 시각화는 코드가 내장되어 있어, 학습 종료 후 Result 폴더에 저장됩니다.

 

참고 문서

[1] 깃허브