A data leakage problem in ADDA method training phase

thuml / Transfer-Learning-Library

Transfer Learning Library for Domain Adaptation, Task Adaptation, and Domain Generalization

http://transfer.thuml.ai

MIT License

3.39k stars 553 forks source link

A data leakage problem in ADDA method training phase #153

Closed Aknifejackzhmolong closed 2 years ago

Aknifejackzhmolong commented 2 years ago

Describe the bug It look like a data leakage problem in files adda.py & utils.py because the val_loader and test_loader is equal some time.

In training phase, the val_loader are used in calculating acc and to save best model, but in test phase the test_loader which are used in calculating accuray performance of model is the same as val_loader.

Expect behavor I want to run a unsupervised domain adaptation program.

Please kindly advise.

Screenshots adda.py

utils.py

thucbx99 commented 2 years ago

Hope issue15, issue28, issue47 and issue105 could answer your question.