Learning to Weighted Time-series via Cascaded Teaching Transformers

This is the origin Pytorch implementation of Cascaded Framework in the following paper: Learning to Weighted Time-series via Cascaded Teaching Transformers

Figure 1. The architecture of cascaded framework.

Requirements

• Python 3.8
• matplotlib == 3.1.1
• numpy == 1.19.4
• pandas == 0.25.1
• scikit_learn == 0.21.3
• torch == 1.10.0

Dependencies can be installed using the following command:

pip install -r requirements.txt

Data

The ETT dataset used in the paper can be download in the repo ETDataset. The required data files should be put into data/ETT/ folder.

Usage

Commands for training and testing the model with Cascaded Framework on Dataset ETTh1:

# ETTh1
python -u main.py --fourrier --pred_len 24 --lambda_par 0.8 --A_lr 0.0002 --A_weight_decay 0 --itr 5\
        --w_weight_decay 0.01 --fourier_divider 40 --temp 5 --name param1 --data ETTh1 --data_path ETTh1.csv
python -u main.py --fourrier --pred_len 48 --lambda_par 0.8--A_lr 0.0002 --A_weight_decay 0 --itr 5\
	--w_weight_decay 0.001 --fourier_divider 40 --temp 1 --name param2 --data ETTh1 --data_path ETTh1.csv
python -u main.py --fourrier --pred_len 168 --lambda_par 0.8 --A_lr 0.0002 --A_weight_decay 0 --itr 5\
	--w_weight_decay 0.0008 --fourier_divider 40 --temp 5 --name param3 --data ETTh1 --data_path ETTh1.csv
python -u main.py --fourrier --pred_len 336 --lambda_par 0.6 --A_lr 0.0002 --A_weight_decay 0 --itr 5\
	--w_weight_decay 0.004 --fourier_divider 40 --temp 5 --name param4 --data ETTh1 --data_path ETTh1.csv
python -u main.py --fourrier --pred_len 720 --lambda_par 0.6 --A_lr 0.0002 --A_weight_decay 0 --itr 5\
	--w_weight_decay 0.0001 --fourier_divider 40 --temp 5 --name param5 --data ETTh1 --data_path ETTh1.csv

We will update all the scripts in the future.

Results

Figure 2. Multivariate forecasting results.