• Priceformer
• • Overview
  • Result on donggua Dataset
  • Dependencies
  • Acknowledge

Priceformer

Priceformer: An algorithm for forecasting commodity price trend based on Transformer

Overview

Here we provide the implementation of a (Priceformer) layer in Pytorch. The repository is organised as follows:

• dataset/ put you data here ;
• models/ contains the implementation of the Priceformer network (Priceformer.py);
• checkpoints/ contains a pre-trained Priceformer model;

Finally, run.py puts all of the above together and may be used to execute a full training run on you data by executing
python3 -u run.py\ --is_training 1\ --root_path ./dataset/Dataset/\ --data_path **Your data path**\ --model_id *Your model ID*\ --model Priceformer\ --data *Your data name*\ --features S\ --seq_len 7\ --label_len 3\ --pred_len 7\ --e_layers 2\ --d_layers 1\ --factor 3\ --enc_in 1\ --dec_in 1\ --c_out 1\ --des 'Exp'\ --itr 1\ --target *Your target* .

Result on donggua Dataset

Dependencies

The script has been tested running under Python 3.5.2, with the following packages installed (along with their dependencies):

• numpy==1.19.2
• scipy==1.0.0
• scikit-learn==0.24.2
• pandas==1.1.5
• matplotlib==3.3.4
• pytorch==1.4.0

In addition, CUDA 9.0 and cuDNN 7 have been used.

Acknowledge

This work was supported by the National Key R&D Program of China under Grant No. 2020AAA0103804(Sponsor: Hefu Liu) and partially supported by grants from the National Natural Science Foundation of China (No.72004021). This work belongs to the University of science and technology of China.