Our experimental environments include Python 3.9, Pytorch 1.13.1 with CUDA 11.6, and transformers 4.31.0. To install all dependencies, please use the below command.
pip install -r requirements.txt
The pre-processed datasets can be obtained from the link here. Then you may choose to download all_datasets.zip, place this zip file into the dataset folder, and finally unzip the file.
In general, we use a csv file to indicate the executing tasks (including training and evaluations) during the cross-domain learning process. There are five columns in the file.
(1) Data: the name of a dataset, corresponding to a config file in the folder data_configs.
(2) Prediction: the prediction length.
(3) Train: the indicator for training.
(4) Valid: the indicator for validation.
(5) Test: the indicator for testing.
For example, the below command is used to train one model for the tasks listed in the file execute_list/train_all.csv.
python run_avg.py --gpu 0 --training_list execute_list/train_all.csv --percent 100
We appreciate the following github repository for sharing the valuable code base and datasets: