Unofficial PyTorch implementation of MELU from the paper: MeLU: Meta-Learned User Preference Estimator for Cold-Start Recommendation.
- python: 3.5+
- Pytorch: 1.0+
We offer the training process and model in multi_result_files/9b8290dd3f63cbafcd141ba21282c783.pkl.
python3 maml.pyYou can modify the detailed parameters according to the definition in maml.py.
python3 maml.py --testBy default, you can directly add the test argument to test the model obtained from the same aruguments setting.
mode_path = utils.get_path_from_args(args)You can also modify the code in maml.py manually since the arguments may vary for training and testing process.
mode_path = '9b8290dd3f63cbafcd141ba21282c783'The official code doesn't offer evaluation code for testing. So based on this implementation, you can test the MAE for 4 partitions. In addition, we find the hyperparameters setting in original paepr isn't reasonable so we modify that and rerun the test.
| Partition | MAE based on our hyperparameters | Reported MAE in original paper |
|---|---|---|
| Existing items for existing users | 0.68$\pm$0.01 | 0.75$\pm $NA |
| Existing items for new users | 0.74$\pm$0.01 | 0.79$\pm$NA |
| new items for existing users | 0.90$\pm$0.01 | 0.92$\pm$NA |
| new items for new users | 0.89$\pm$0.02 | 0.92$\pm$NA |
This code refers code from: wyharveychen/CloserLookFewShot. lmzintgraf/cavia. hoyeoplee/MeLU.