Predict and interpret adaptation to data class (#586)

* adapt predict and interpret to config classes

clinicadl/config/config/__init__.py

@@ -5,7 +5,6 @@
 from .data import DataConfig
 from .dataloader import DataLoaderConfig
 from .early_stopping import EarlyStoppingConfig
-from .interpret import InterpretConfig
 from .lr_scheduler import LRschedulerConfig
 from .maps_manager import MapsManagerConfig
 from .modality import (
@@ -17,7 +16,6 @@
 from .model import ModelConfig
 from .optimization import OptimizationConfig
 from .optimizer import OptimizerConfig
-from .predict import PredictConfig
 from .preprocessing import (
     PreprocessingConfig,
     PreprocessingImageConfig,

clinicadl/config/config/computational.py

@@ -1,6 +1,10 @@
 from logging import getLogger
 
-from pydantic import BaseModel, ConfigDict
+from pydantic import BaseModel, ConfigDict, model_validator
+from typing_extensions import Self
+
+from clinicadl.utils.cmdline_utils import check_gpu
+from clinicadl.utils.exceptions import ClinicaDLArgumentError
 
 logger = getLogger("clinicadl.computational_config")
 
@@ -13,3 +17,13 @@ class ComputationalConfig(BaseModel):
     gpu: bool = True
     # pydantic config
     model_config = ConfigDict(validate_assignment=True)
+
+    @model_validator(mode="after")
+    def validator_gpu(self) -> Self:
+        if self.gpu:
+            check_gpu()
+        elif self.amp:
+            raise ClinicaDLArgumentError(
+                "AMP is designed to work with modern GPUs. Please add the --gpu flag."
+            )
+        return self

clinicadl/config/config/cross_validation.py

@@ -5,6 +5,8 @@
 from pydantic import BaseModel, ConfigDict, field_validator
 from pydantic.types import NonNegativeInt
 
+from clinicadl.utils.maps_manager.maps_manager import MapsManager
+
 logger = getLogger("clinicadl.cross_validation_config")
 
 
@@ -19,7 +21,7 @@ class CrossValidationConfig(
 
     n_splits: NonNegativeInt = 0
     split: Optional[Tuple[NonNegativeInt, ...]] = None
-    tsv_directory: Path
+    tsv_directory: Optional[Path] = None  # not needed in predict ?
     # pydantic config
     model_config = ConfigDict(validate_assignment=True)
 
@@ -28,3 +30,9 @@ def validator_split(cls, v):
         if isinstance(v, list):
             return tuple(v)
         return v  # TODO : check that split exists (and check coherence with n_splits)
+
+    def adapt_cross_val_with_maps_manager_info(self, maps_manager: MapsManager):
+        # TEMPORARY
+        if not self.split:
+            self.split = maps_manager._find_splits()
+        logger.debug(f"List of splits {self.split}")

clinicadl/config/config/data.py

@@ -6,6 +6,7 @@
 
 from clinicadl.utils.caps_dataset.data import load_data_test
 from clinicadl.utils.enum import Mode
+from clinicadl.utils.maps_manager.maps_manager import MapsManager
 from clinicadl.utils.preprocessing import read_preprocessing
 
 logger = getLogger("clinicadl.data_config")
@@ -24,12 +25,17 @@ class DataConfig(BaseModel):  # TODO : put in data module
     label: Optional[str] = None
     label_code: Dict[str, int] = {}
     multi_cohort: bool = False
-    preprocessing_json: Path
+    preprocessing_json: Optional[Path] = None
     data_tsv: Optional[Path] = None
     n_subjects: int = 300
     # pydantic config
     model_config = ConfigDict(validate_assignment=True)
 
+    def adapt_data_with_maps_manager_info(self, maps_manager: MapsManager):
+        # TEMPORARY
+        if self.diagnoses is None or len(self.diagnoses) == 0:
+            self.diagnoses = maps_manager.diagnoses
+
     def create_groupe_df(self):
         group_df = None
         if self.data_tsv is not None and self.data_tsv.is_file():

clinicadl/config/config/dataloader.py

@@ -4,6 +4,7 @@
 from pydantic.types import PositiveInt
 
 from clinicadl.utils.enum import Sampler
+from clinicadl.utils.maps_manager.maps_manager import MapsManager
 
 logger = getLogger("clinicadl.dataloader_config")
 
@@ -16,3 +17,11 @@ class DataLoaderConfig(BaseModel):  # TODO : put in data/splitter module
     sampler: Sampler = Sampler.RANDOM
     # pydantic config
     model_config = ConfigDict(validate_assignment=True)
+
+    def adapt_dataloader_with_maps_manager_info(self, maps_manager: MapsManager):
+        # TEMPORARY
+        if not self.batch_size:
+            self.batch_size = maps_manager.batch_size
+
+        if not self.n_proc:
+            self.n_proc = maps_manager.n_proc

clinicadl/config/config/maps_manager.py

@@ -26,17 +26,3 @@ def check_output_saving_nifti(self, network_task: str) -> None:
             raise ClinicaDLArgumentError(
                 "Cannot save nifti if the network task is not reconstruction. Please remove --save_nifti option."
             )
-
-    def adapt_config_with_maps_manager_info(self, maps_manager: MapsManager):
-        if not self.split_list:
-            self.split_list = maps_manager._find_splits()
-        logger.debug(f"List of splits {self.split_list}")
-
-        if self.diagnoses is None or len(self.diagnoses) == 0:
-            self.diagnoses = maps_manager.diagnoses
-
-        if not self.batch_size:
-            self.batch_size = maps_manager.batch_size
-
-        if not self.n_proc:
-            self.n_proc = maps_manager.n_proc

clinicadl/config/config/interpret.py → clinicadl/config/config/pipelines/interpret.py

@@ -5,18 +5,28 @@
 
 from pydantic import BaseModel, field_validator
 
+from clinicadl.config.config import (
+    ComputationalConfig,
+    CrossValidationConfig,
+    DataLoaderConfig,
+    MapsManagerConfig,
+    ValidationConfig,
+)
+from clinicadl.config.config import DataConfig as DataBaseConfig
 from clinicadl.interpret.gradients import GradCam, Gradients, VanillaBackProp
 from clinicadl.utils.caps_dataset.data import (
     load_data_test,
 )
 from clinicadl.utils.enum import InterpretationMethod
-from clinicadl.utils.exceptions import ClinicaDLArgumentError  # type: ignore
-from clinicadl.utils.maps_manager.maps_manager import MapsManager  # type: ignore
 
-logger = getLogger("clinicadl.predict_config")
+logger = getLogger("clinicadl.interpret_config")
+
+
+class DataConfig(DataBaseConfig):
+    caps_directory: Optional[Path] = None
 
 
-class InterpretConfig(BaseModel):
+class InterpretBaseConfig(BaseModel):
     name: str
     method: InterpretationMethod = InterpretationMethod.GRADIENTS
     target_node: int = 0
@@ -38,3 +48,15 @@ def get_method(self) -> Gradients:
             return GradCam
         else:
             raise ValueError(f"The method {self.method.value} is not implemented")
+
+
+class InterpretConfig(
+    MapsManagerConfig,
+    InterpretBaseConfig,
+    DataConfig,
+    ValidationConfig,
+    CrossValidationConfig,
+    ComputationalConfig,
+    DataLoaderConfig,
+):
+    """Config class to perform Transfer Learning."""

clinicadl/config/config/predict.py → clinicadl/config/config/pipelines/predict.py

@@ -2,19 +2,42 @@
 
 from pydantic import BaseModel
 
+from clinicadl.config.config.data import DataConfig as DataBaseConfig
+from clinicadl.config.config.maps_manager import (
+    MapsManagerConfig as MapsManagerBaseConfig,
+)
 from clinicadl.utils.exceptions import ClinicaDLArgumentError  # type: ignore
 
+from ..computational import ComputationalConfig
+from ..cross_validation import CrossValidationConfig
+from ..dataloader import DataLoaderConfig
+from ..validation import ValidationConfig
+
 logger = getLogger("clinicadl.predict_config")
 
 
-class PredictConfig(BaseModel):
+class MapsManagerConfig(MapsManagerBaseConfig):
     save_tensor: bool = False
     save_latent_tensor: bool = False
-    use_labels: bool = True
 
     def check_output_saving_tensor(self, network_task: str) -> None:
         # Check if task is reconstruction for "save_tensor" and "save_nifti"
         if self.save_tensor and network_task != "reconstruction":
             raise ClinicaDLArgumentError(
                 "Cannot save tensors if the network task is not reconstruction. Please remove --save_tensor option."
             )
+
+
+class DataConfig(DataBaseConfig):
+    use_labels: bool = True
+
+
+class PredictConfig(
+    MapsManagerConfig,
+    DataConfig,
+    ValidationConfig,
+    CrossValidationConfig,
+    ComputationalConfig,
+    DataLoaderConfig,
+):
+    """Config class to perform Transfer Learning."""

clinicadl/train/tasks/classification/config.py → clinicadl/config/config/pipelines/task/classification.py

@@ -6,7 +6,7 @@
 from clinicadl.config.config import DataConfig as BaseDataConfig
 from clinicadl.config.config import ModelConfig as BaseModelConfig
 from clinicadl.config.config import ValidationConfig as BaseValidationConfig
-from clinicadl.train.trainer.training_config import TrainingConfig
+from clinicadl.config.config.pipelines.train import TrainConfig
 from clinicadl.utils.enum import ClassificationLoss, ClassificationMetric, Task
 
 logger = getLogger("clinicadl.classification_config")
@@ -57,7 +57,7 @@ def list_to_tuples(cls, v):
         return v
 
 
-class ClassificationConfig(TrainingConfig):
+class ClassificationConfig(TrainConfig):
     """
     Config class for the training of a classification model.
 

clinicadl/train/tasks/reconstruction/config.py → clinicadl/config/config/pipelines/task/reconstruction.py

@@ -6,7 +6,7 @@
 
 from clinicadl.config.config import ModelConfig as BaseModelConfig
 from clinicadl.config.config import ValidationConfig as BaseValidationConfig
-from clinicadl.train.trainer.training_config import TrainingConfig
+from clinicadl.config.config.pipelines.train import TrainConfig
 from clinicadl.utils.enum import (
     Normalization,
     ReconstructionLoss,
@@ -47,7 +47,7 @@ def list_to_tuples(cls, v):
         return v
 
 
-class ReconstructionConfig(TrainingConfig):
+class ReconstructionConfig(TrainConfig):
     """
     Config class for the training of a reconstruction model.
 

clinicadl/train/tasks/regression/config.py → clinicadl/config/config/pipelines/task/regression.py

@@ -7,7 +7,7 @@
 from clinicadl.config.config import DataConfig as BaseDataConfig
 from clinicadl.config.config import ModelConfig as BaseModelConfig
 from clinicadl.config.config import ValidationConfig as BaseValidationConfig
-from clinicadl.train.trainer.training_config import TrainingConfig
+from clinicadl.config.config.pipelines.train import TrainConfig
 from clinicadl.utils.enum import RegressionLoss, RegressionMetric, Task
 
 logger = getLogger("clinicadl.reconstruction_config")
@@ -47,7 +47,7 @@ def list_to_tuples(cls, v):
         return v
 
 
-class RegressionConfig(TrainingConfig):
+class RegressionConfig(TrainConfig):
     """
     Config class for the training of a regression model.
 

clinicadl/train/trainer/training_config.py → clinicadl/config/config/pipelines/train.py

@@ -30,7 +30,7 @@
 logger = getLogger("clinicadl.training_config")
 
 
-class TrainingConfig(BaseModel, ABC):
+class TrainConfig(BaseModel, ABC):
     """
 
     Abstract config class for the training pipeline.

clinicadl/config/options/__init__.py

@@ -1 +1 @@
-from .task import classification, reconstruction, regression
+# from .task import classification, reconstruction, regression

clinicadl/config/options/callbacks.py

@@ -1,21 +1,20 @@
 import click
 
-import clinicadl.train.trainer.training_config as config
-from clinicadl.config import config
+from clinicadl.config.config.callbacks import CallbacksConfig
 from clinicadl.utils.config_utils import get_default_from_config_class as get_default
 from clinicadl.utils.config_utils import get_type_from_config_class as get_type
 
 emissions_calculator = click.option(
     "--calculate_emissions/--dont_calculate_emissions",
-    default=get_default("emissions_calculator", config.CallbacksConfig),
+    default=get_default("emissions_calculator", CallbacksConfig),
     help="Flag to allow calculate the carbon emissions during training.",
     show_default=True,
 )
 track_exp = click.option(
     "--track_exp",
     "-te",
-    type=get_type("track_exp", config.CallbacksConfig),
-    default=get_default("track_exp", config.CallbacksConfig),
+    type=get_type("track_exp", CallbacksConfig),
+    default=get_default("track_exp", CallbacksConfig),
     help="Use `--track_exp` to enable wandb/mlflow to track the metric (loss, accuracy, etc...) during the training.",
     show_default=True,
 )

clinicadl/config/options/computational.py

@@ -1,13 +1,13 @@
 import click
 
-from clinicadl.config import config
+from clinicadl.config.config.computational import ComputationalConfig
 from clinicadl.utils.config_utils import get_default_from_config_class as get_default
 from clinicadl.utils.config_utils import get_type_from_config_class as get_type
 
 # Computational
 amp = click.option(
     "--amp/--no-amp",
-    default=get_default("amp", config.ComputationalConfig),
+    default=get_default("amp", ComputationalConfig),
     help="Enables automatic mixed precision during training and inference.",
     show_default=True,
 )
@@ -21,7 +21,7 @@
 )
 gpu = click.option(
     "--gpu/--no-gpu",
-    default=get_default("gpu", config.ComputationalConfig),
+    default=get_default("gpu", ComputationalConfig),
     help="Use GPU by default. Please specify `--no-gpu` to force using CPU.",
     show_default=True,
 )

clinicadl/config/options/cross_validation.py

@@ -1,15 +1,14 @@
 import click
 
-import clinicadl.train.trainer.training_config as config
-from clinicadl.config import config
+from clinicadl.config.config.cross_validation import CrossValidationConfig
 from clinicadl.utils.config_utils import get_default_from_config_class as get_default
 from clinicadl.utils.config_utils import get_type_from_config_class as get_type
 
 # Cross Validation
 n_splits = click.option(
     "--n_splits",
-    type=get_type("n_splits", config.CrossValidationConfig),
-    default=get_default("n_splits", config.CrossValidationConfig),
+    type=get_type("n_splits", CrossValidationConfig),
+    default=get_default("n_splits", CrossValidationConfig),
     help="If a value is given for k will load data of a k-fold CV. "
     "Default value (0) will load a single split.",
     show_default=True,
@@ -18,7 +17,7 @@
     "--split",
     "-s",
     type=int,  # get_type("split", config.CrossValidationConfig),
-    default=get_default("split", config.CrossValidationConfig),
+    default=get_default("split", CrossValidationConfig),
     multiple=True,
     help="Train the list of given splits. By default, all the splits are trained.",
     show_default=True,