README.md

Spatio-temporal Continual Learning Setting

In this sub-setting, we use and adapt the code repository of the EAC framework.

📚 Training Data

The processed dataset can be directly accessed from the cloud disk!

Please download all processed datasets and place them in the data folder.

🚀 Getting Started

Installation

1. Please install the core dependencies, including:

python = 3.8.5
pytorch = 1.7.1
torch-geometric = 1.6.3

2. Or you can directly create and import a ready-made environment:

conda env create -f environment.yaml
conda activate stg

Usages

Before you get started, you need to make sure your data and trained weights are ready!

---

To test EAC or STKEC w/o ST-TTC (i.e., Normal test) on different datasets:

First, you should make sure that line 60 & line 153 in main.py file and line 56 & line 148 in stkec_main.py contains the following:

test_model(model, args, test_loader, pin_memory=True)

and then, your can run a specific method on a specific dataset separately:

python main.py --conf conf/PEMS/eac.json --gpuid 0 --seed 43
python stkec_main.py --conf conf/PEMS/eac.json --gpuid 0 --seed 43

Or you can run the script to batch execute all baseline methods on a specified dataset, for example, run all baseline methods on the PEMS-Stream dataset:

sh scripts/pems_run.sh

---

To test EAC or STKEC w/ ST-TTC on different datasets:

First, you should make sure that line 60 & line 153 in main.py file and line 56 & line 148 in stkec_main.py contains the following:

test_model_with_ttc(model, args, test_loader, pin_memory=True)

and then, your can run a specific method on a specific dataset separately:

python main.py --conf conf/PEMS/eac.json --gpuid 0 --seed 43
python stkec_main.py --conf conf/PEMS/eac.json --gpuid 0 --seed 43

Or you can run the script to batch execute all baseline methods on a specified dataset, for example, run all baseline methods on the PEMS-Stream dataset:

sh scripts/pems_run.sh