Thibault de Varax

Don’t we need to add some methods in `EarlyStopping`? I think it inherits from `Metrics` for the `check_metrics` methods _Originally posted by @camillebrianceau in https://github.com/aramis-lab/clinicadl/pull/796#discussion_r2332576249_

Create new MAPS

1

- [x] integrate ``torchsummary`` - [x] create a ``summary.log`` - [x] create a ``architecture.log`` - [x] add main file's path in ``summary.log`` - [x] training time and max memory consumption...

Refactor Unittests

1

Currently, unittests are not really independent from each other. Besides, some improvements are needed, e.g. use pytest's ``tmp_path`` fixture.

- [ ] Gradient Clipping - [ ] Channel Last Memory Format - [ ] AMP - [ ] DDP - [ ] FSDP - [ ] Cluster compatibility

Currently, ``TensorConversion`` and ``CapsDataset`` are not independent at all. My opinion is that they should both dependant on ``CapsReader``, but must be independent of each other.

Many unittests create manually their own temporary directory, whereas they could use pytest ``tmp_path`` fixture.