Train all parameters or only BN layers

As I understand from the paper, you intend to only train the params of the BN layers. Is this what you are trying to do at [this line](https://github.com/bethgelab/robustness/blob/0ef82b178e3526f63d8017dc870072d611d883e9/examples/selflearning/main.py#L92)? If so please be aware that module.eval() does not freeze the module’s params. If this is not what you are trying to do please kindly let me know what is the purpose of the mentioned line.

Thanks

Read examples/selflearning/main.py at line 92 and compare the intended training behavior with the paper. Check how module.eval() relates to parameter training in this path; done means the issue’s question is answered clearly and the implementation matches the intended scope, if a change is required.

Written by the indexing model from the issue text.