remove SSDA

clinicadl/commandline/pipelines/train/classification/cli.py

@@ -13,7 +13,6 @@
     optimizer,
     reproducibility,
     split,
-    ssda,
     transforms,
     validation,
 )
@@ -63,12 +62,6 @@
 @dataloader.batch_size
 @dataloader.sampler
 @dataloader.n_proc
-# ssda option
-@ssda.ssda_network
-@ssda.caps_target
-@ssda.tsv_target_lab
-@ssda.tsv_target_unlab
-@ssda.preprocessing_json_target
 # Cross validation
 @split.n_splits
 @split.split

clinicadl/commandline/pipelines/train/reconstruction/cli.py

@@ -13,7 +13,6 @@
     optimizer,
     reproducibility,
     split,
-    ssda,
     transforms,
     validation,
 )
@@ -63,12 +62,6 @@
 @dataloader.batch_size
 @dataloader.sampler
 @dataloader.n_proc
-# ssda option
-@ssda.ssda_network
-@ssda.caps_target
-@ssda.tsv_target_lab
-@ssda.tsv_target_unlab
-@ssda.preprocessing_json_target
 # Cross validation
 @split.n_splits
 @split.split

clinicadl/commandline/pipelines/train/regression/cli.py

@@ -13,7 +13,6 @@
     optimizer,
     reproducibility,
     split,
-    ssda,
     transforms,
     validation,
 )
@@ -61,12 +60,6 @@
 @dataloader.batch_size
 @dataloader.sampler
 @dataloader.n_proc
-# ssda o
-@ssda.ssda_network
-@ssda.caps_target
-@ssda.tsv_target_lab
-@ssda.tsv_target_unlab
-@ssda.preprocessing_json_target
 # Cross validation
 @split.n_splits
 @split.split

clinicadl/nn/networks/__init__.py

@@ -8,7 +8,6 @@
     resnet18,
 )
 from .random import RandomArchitecture
-from .ssda import Conv5_FC3_SSDA
 from .unet import UNet
 from .vae import (
     CVAE_3D,

clinicadl/predictor/predictor.py

@@ -1044,85 +1044,6 @@ def _test_loader(
                     data_group=data_group,
                 )
 
-    def _test_loader_ssda(
-        self,
-        maps_manager: MapsManager,
-        dataloader,
-        criterion,
-        alpha,
-        data_group,
-        split,
-        selection_metrics,
-        use_labels=True,
-        gpu=None,
-        network=None,
-        target=False,
-        report_ci=True,
-    ):
-        """
-        Launches the testing task on a dataset wrapped by a DataLoader and writes prediction TSV files.
-
-        Args:
-            dataloader (torch.utils.data.DataLoader): DataLoader wrapping the test CapsDataset.
-            criterion (torch.nn.modules.loss._Loss): optimization criterion used during training.
-            data_group (str): name of the data group used for the testing task.
-            split (int): Index of the split used to train the model tested.
-            selection_metrics (list[str]): List of metrics used to select the best models which are tested.
-            use_labels (bool): If True, the labels must exist in test meta-data and metrics are computed.
-            gpu (bool): If given, a new value for the device of the model will be computed.
-            network (int): Index of the network tested (only used in multi-network setting).
-        """
-        for selection_metric in selection_metrics:
-            log_dir = (
-                maps_manager.maps_path
-                / f"split-{split}"
-                / f"best-{selection_metric}"
-                / data_group
-            )
-            maps_manager.write_description_log(
-                log_dir,
-                data_group,
-                dataloader.dataset.caps_dict,
-                dataloader.dataset.df,
-            )
-
-            # load the best trained model during the training
-            model, _ = maps_manager._init_model(
-                transfer_path=maps_manager.maps_path,
-                split=split,
-                transfer_selection=selection_metric,
-                gpu=gpu,
-                network=network,
-            )
-            prediction_df, metrics = self.test_da(
-                network_task=maps_manager.network_task,
-                model=model,
-                dataloader=dataloader,
-                criterion=criterion,
-                target=target,
-                report_ci=report_ci,
-                mode=maps_manager.mode,
-                metrics_module=maps_manager.metrics_module,
-                n_classes=maps_manager.n_classes,
-            )
-            if use_labels:
-                if network is not None:
-                    metrics[f"{maps_manager.mode}_id"] = network
-
-                if report_ci:
-                    loss_to_log = metrics["Metric_values"][-1]
-                else:
-                    loss_to_log = metrics["loss"]
-
-                logger.info(
-                    f"{maps_manager.mode} level {data_group} loss is {loss_to_log} for model selected on {selection_metric}"
-                )
-
-            # Replace here
-            maps_manager._mode_level_to_tsv(
-                prediction_df, metrics, split, selection_metric, data_group=data_group
-            )
-
     @torch.no_grad()
     def _compute_output_tensors(
         self,

clinicadl/random_search/random_search_utils.py

@@ -124,7 +124,6 @@ def random_sampling(rs_options: Dict[str, Any]) -> Dict[str, Any]:
         "mode": "fixed",
         "multi_cohort": "fixed",
         "multi_network": "choice",
-        "ssda_netork": "fixed",
         "n_fcblocks": "randint",
         "n_splits": "fixed",
         "n_proc": "fixed",

clinicadl/resources/config/train_config.toml

@@ -4,7 +4,6 @@
 [Model]
 architecture = "default" # ex : Conv5_FC3
 multi_network = false
-ssda_network = false
 
 [Architecture]
 # CNN

clinicadl/trainer/config/train.py

@@ -14,7 +14,6 @@
 from clinicadl.caps_dataset.dataloader_config import DataLoaderConfig
 from clinicadl.config.config.lr_scheduler import LRschedulerConfig
 from clinicadl.config.config.reproducibility import ReproducibilityConfig
-from clinicadl.config.config.ssda import SSDAConfig
 from clinicadl.maps_manager.config import MapsManagerConfig
 from clinicadl.network.config import NetworkConfig
 from clinicadl.optimizer.optimization import OptimizationConfig
@@ -50,7 +49,6 @@ class TrainConfig(BaseModel, ABC):
     optimizer: OptimizerConfig
     reproducibility: ReproducibilityConfig
     split: SplitConfig
-    ssda: SSDAConfig
     transfer_learning: TransferLearningConfig
     transforms: TransformsConfig
     validation: ValidationConfig
@@ -77,7 +75,6 @@ def __init__(self, **kwargs):
             optimizer=kwargs,
             reproducibility=kwargs,
             split=kwargs,
-            ssda=kwargs,
             transfer_learning=kwargs,
             transforms=kwargs,
             validation=kwargs,
@@ -97,7 +94,6 @@ def _update(self, config_dict: Dict[str, Any]) -> None:
         self.optimizer.__dict__.update(config_dict)
         self.reproducibility.__dict__.update(config_dict)
         self.split.__dict__.update(config_dict)
-        self.ssda.__dict__.update(config_dict)
         self.transfer_learning.__dict__.update(config_dict)
         self.transforms.__dict__.update(config_dict)
         self.validation.__dict__.update(config_dict)