hank0626 / pdf

An official implementation of "Periodicity Decoupling Framework for Long-term Series Forecasting" (ICLR 2024)

Periodicity Decoupling Framework for Long-term Series Forecasting 🚀

[Paper Page] [知乎中文解读] [时序人中文解读]

📰 News

• [2024-01-15] 🎉 Our paper has been accepted by ICLR 2024.

🌟 Results

Quantitatively, compared with Transformer-based models, PDF(720) yields an overall 14.59% reduction in MSE and 10.77% reduction in MAE. Compared with CNN-based models, PDF(720) yields an overall 24.61% reduction in MSE and 19.91% reduction in MAE. Compared with Linear-based models, PDF(720) yields an overall 7.05% reduction in MSE and 5.51% reduction in MAE.

🛠 Prerequisites

Ensure you are using Python 3.9 and install the necessary dependencies by running:

pip install -r requirements.txt

📊 Prepare Datastes

Begin by downloading the required datasets. All datasets are conveniently available at Autoformer. Create a separate folder named ./dataset and neatly organize all the csv files as shown below:

dataset
└── electricity.csv
└── ETTh1.csv
└── ETTh2.csv
└── ETTm1.csv
└── ETTm2.csv
└── traffic.csv
└──  weather.csv

💻 Training

All scripts are located in ./scripts/PDF. Choose from two historical input lengths: 336 and 720. For instance, to train a model using the ETTh2 dataset with an input length of 720, simply run:

sh ./scripts/PDF/720/ETTh2.sh

After training:

• Your trained model will be safely stored in ./checkpoints.
• Visualization outputs are available in ./test_results.
• Numerical results in .npy format can be found in ./results.
• A comprehensive summary of quantitative metrics is accessible in ./results.txt.

📚 Citation

If you find this repo useful, please consider citing our paper as follows:

@article{dai2024period,
  title={Periodicity Decoupling Framework for Long-term Series Forecasting},
  author={Dai, Tao and Wu, Beiliang and Liu, Peiyuan and Li, Naiqi and Bao, Jigang and Jiang, Yong and Xia, Shu-Tao},
  journal={International Conference on Learning Representations},
  year={2024}
}

🙏 Acknowledgement

Special thanks to the following repositories for their invaluable code and datasets:

• https://github.com/thuml/Autoformer
• https://github.com/yuqinie98/PatchTST
• https://github.com/Thinklab-SJTU/Crossformer
• https://github.com/wanghq21/MICN
• https://github.com/thuml/Time-Series-Library
• https://github.com/MAZiqing/FEDformer
• https://github.com/zhouhaoyi/Informer2020

📩 Contact

If you have any questions, please contact [email protected] or submit an issue.

pdf's Issues

代码报错

D:\anaconda\envs\py38\python.exe H:\study\data2\PDF-main\run_longExp.py
Traceback (most recent call last):
File "H:\study\data2\PDF-main\run_longExp.py", line 159, in
exp = Exp(args) # set experiments
File "H:\study\data2\PDF-main\exp\exp_main.py", line 27, in init
super(Exp_Main, self).init(args)
File "H:\study\data2\PDF-main\exp\exp_basic.py", line 10, in init
self.model = self._build_model().to(self.device)
File "H:\study\data2\PDF-main\exp\exp_main.py", line 40, in _build_model
model = model_dict[self.args.model].Model(self.args).float()
File "H:\study\data2\PDF-main\models\PDF.py", line 53, in init
self.model = PDF_backbone(c_in=c_in, context_window=context_window, target_window=target_window,
File "H:\study\data2\PDF-main\layers\PDF_backbone.py", line 36, in init
self.kernel_list = [(n, patch_len[i]) for i, n in enumerate(self.period_len)]
File "H:\study\data2\PDF-main\layers\PDF_backbone.py", line 36, in
self.kernel_list = [(n, patch_len[i]) for i, n in enumerate(self.period_len)]
IndexError: list index out of range
Args in experiment:
Namespace(activation='gelu', add=False, affine=0, attn_dropout=0.05, batch_size=16, c_out=7, checkpoints='./checkpoints/', d_ff=2048, d_layers=1, d_model=512, data='ETT-small', data_path='ETTh1.csv', dec_in=7, decomposition=0, des='test', devices='0', distil=True, do_predict=False, dropout=0.05, e_layers=2, embed='timeF', embed_type=0, enc_in=7, factor=1, fc_dropout=0.0, features='M', freq='h', gpu=0, head_dropout=0.0, individual=0, is_training=1, itr=2, kernel_list=[3, 7, 9], kernel_size=25, label_len=48, learning_rate=0.0001, log='./logs/LongForecasting/PatchTST_Electricity_336_96.log', loss='mse', lradj='type3', model='PDF', model_id='test', moving_avg=25, n_heads=8, num_workers=10, output_attention=False, padding_patch='end', patch_len=[16], patience=100, pct_start=0.3, period=[24, 12], pred_len=96, random_seed=2021, revin=1, root_path='./dataset/ETT-small/', seq_len=96, serial_conv=False, stride=None, subtract_last=0, target='OT', test_flop=False, train_epochs=100, use_amp=False, use_gpu=True, use_multi_gpu=False, wo_conv=False)
Use GPU: cuda:0

您好，代码一直报错，运行不了，麻烦能看一下吗？

help，Please

May I ask what problems this model mainly solves in time series forecasting? This is what my teacher asked me. I am not sure. Please help.

May I ask why your core code Multi-periodic Decoupling Block is not reflected in the PDF?

kernel_list的选取与周期的定义

您好，作者大大，在代码中没看到确定在周期的代码呢？还有一个问题kernel_list的大小是怎么确定的啊？可以麻烦您解答一下吗？谢谢

frequency selection

Hello, I would like to ask where the "frequency selection" mentioned in the paper is in the code? I didn't find it in the official code implementation

Period Patching

请问Period Patching中的p是怎么确定的？

特定数据集超参数的选取策略

您好，请问一下模型中几个主要参数 period、patch_len、stride 有大致的选取原则吗？比如傅里叶变换的高频是如何跟参数对应的，应该选择几个参数，参数之间有没有相关的关系。

predict_merge

Hello，when I use PDF, an error come up here.

How to define the "predict_merge" in model（Autoformer, Informer and Transformer）?

We need to use "Parser" to add the argument — "predict_merge" firstly before using PDF,right?

代码问题

if [ ! -d "./logs" ]; then
mkdir ./logs
fi

if [ ! -d "./logs/LongForecasting" ]; then
mkdir ./logs/LongForecasting
fi

model_name=PDF
root_path_name=./dataset/ETT-small
data_path_name=ETTh1.csv
model_id_name=ETTh1
data_name=ETTh1

random_seed=2021

seq_len=720
for pred_len in 96 192
do
python -u run_longExp.py
--random_seed 2021
--is_training 1
--root_path ./dataset/
--data_path ETTh1.csv
--model_id ETTh1''$seq_len''$pred_len
--model PDF
--data ETTh1
--features M
--seq_len $seq_len
--pred_len $pred_len
--enc_in 7
--e_layers 3
--n_heads 4
--d_model 16
--d_ff 128
--dropout 0.25
--fc_dropout 0.15
--kernel_list 3 7 11
--period 24
--patch_len 1
--stride 1
--des Exp
--pct_start 0.2
--train_epochs 100
--patience 10
--itr 1 --batch_size 128 --learning_rate 0.0001 >logs/LongForecasting/$model_name''$model_id_name''$seq_len'_'$pred_len.log
done

for pred_len in 336 720
do
python -u run_longExp.py
--random_seed 2021
--is_training 1
--root_path ./dataset/
--data_path ETTh1.csv
--model_id ETTh1''$seq_len''$pred_len
--model PDF
--data ETTh1
--features M
--seq_len $seq_len
--pred_len $pred_len
--enc_in 7
--e_layers 3
--n_heads 4
--d_model 16
--d_ff 128
--dropout 0.5
--fc_dropout 0.25
--kernel_list 3 7 11
--period 24
--patch_len 1
--stride 1
--des Exp
--pct_start 0.2
--train_epochs 100
--patience 10
--itr 1 --batch_size 128 --learning_rate 0.00025 >logs/LongForecasting/$model_name''$model_id_name''$seq_len'_'$pred_len.log
done

您好运行这个.sh文件后报如下错误：D:\anaconda\envs\py38\python.exe: can't open file 'run_longExp.py': [Errno 2] No such file or directory 需要把这个run_longExp.py文件放到哪里吗？
麻烦您了