PyTorch Implementation for "Temporal Domain Generalization via Learning Instance-level Evolving Patterns"
This example code include both classificaiton and regression tasks. To run our experiments on 2-Moons dataset, go to the classification folder. To run our experiments on House dataset, go to the regression folder.
1.Pretraining
bash scripts/pretrain.sh
We have also provided our pre-trained models in the models folder.
2.Generate trajectory data (Instance Evolving Trajectory Mining)
bash scripts/gen_tra_data.sh
We have also provided our generated trajectory data in the data folder.
3.Train the continuous-time model and make predictions for the target domain
bash scripts/train.sh
Environments used in our experiments:
- Python 3.8.18
- PyTorch 1.9.1
- Numpy 1.24.3
- torchvision 0.9.1
- POT 0.9.1
- torchdiffeq 0.2.3
- torchsde 0.2.6