First draft for KFold (#684)

* make_split * make_kfold * KFold *SingleSplit --------- Co-authored-by: camillebrianceau <[email protected]> Co-authored-by: camillebrianceau <[email protected]>
aramis-lab · Dec 18, 2024 · 621cc96 · 621cc96
1 parent 464dddf
commit 621cc96
Show file tree

Hide file tree

Showing 54 changed files with 2,923 additions and 546 deletions.
diff --git a/clinicadl/API/complicated_case.py b/clinicadl/API/complicated_case.py
@@ -2,19 +2,15 @@
 
 import torchio.transforms as transforms
 
-from clinicadl.dataset.caps_dataset import (
-    CapsDatasetPatch,
-    CapsDatasetRoi,
-    CapsDatasetSlice,
-)
-from clinicadl.dataset.caps_reader import CapsReader
-from clinicadl.dataset.concat import ConcatDataset
-from clinicadl.dataset.config.extraction import ExtractionConfig
-from clinicadl.dataset.config.preprocessing import (
-    PreprocessingConfig,
-    T1PreprocessingConfig,
-)
 from clinicadl.dataset.dataloader_config import DataLoaderConfig
+from clinicadl.dataset.datasets.caps_dataset import CapsDataset
+from clinicadl.dataset.datasets.concat import ConcatDataset
+from clinicadl.dataset.preprocessing import (
+    PreprocessingCustom,
+    PreprocessingPET,
+    PreprocessingT1,
+)
+from clinicadl.dataset.readers.caps_reader import CapsReader
 from clinicadl.experiment_manager.experiment_manager import ExperimentManager
 from clinicadl.losses.config import CrossEntropyLossConfig
 from clinicadl.losses.factory import get_loss_function
@@ -31,6 +27,7 @@
 from clinicadl.splitter.kfold import KFolder
 from clinicadl.splitter.split import get_single_split, split_tsv
 from clinicadl.trainer.trainer import Trainer
+from clinicadl.transforms.extraction import ROI, BaseExtraction, Image, Patch, Slice
 from clinicadl.transforms.transforms import Transforms
 
 # Create the Maps Manager / Read/write manager /
@@ -40,60 +37,47 @@
 )  # a ajouter dans le manager: mlflow/ profiler/ etc ...
 
 caps_directory = Path("caps_directory")  # output of clinica pipelines
-caps_reader = CapsReader(caps_directory, manager=manager)
-
-preprocessing_1 = caps_reader.get_preprocessing("t1-linear")
-caps_reader.prepare_data(
-    preprocessing=preprocessing_1, data_tsv=Path(""), n_proc=2
-)  # don't return anything -> just extract the image tensor and compute some information for each images
-
-transforms_1 = Transforms(
-    object_augmentation=[transforms.Crop, transforms.Transform],
-    image_augmentation=[transforms.Crop, transforms.Transform],
-    extraction=ExtractionPatchConfig(patch_size=3),
-    image_transforms=[transforms.Blur, transforms.Ghosting],
-    object_transforms=[transforms.BiasField, transforms.Motion],
-)  # not mandatory
-
-preprocessing_2 = caps_reader.get_preprocessing("pet-linear")
-caps_reader.prepare_data(
-    preprocessing=preprocessing_1, data_tsv=Path(""), n_proc=2
-)  # to extract the tensor of the PET file this time
-transforms_2 = Transforms(
-    object_augmentation=[transforms.Crop, transforms.Transform],
-    image_augmentation=[transforms.Crop, transforms.Transform],
-    extraction=ExtractionSliceConfig(),
-    image_transforms=[transforms.Blur, transforms.Ghosting],
-    object_transforms=[transforms.BiasField, transforms.Motion],
+
+sub_ses_t1 = Path("/Users/camille.brianceau/aramis/CLINICADL/caps/subjects_t1.tsv")
+preprocessing_t1 = PreprocessingT1()
+transforms_image = Transforms(
+    image_augmentation=[transforms.RandomMotion()],
+    extraction=Image(),
+    image_transforms=[transforms.Blur((0.5, 0.6, 0.3))],
+)
+
+print("T1 and image ")
+
+dataset_t1_image = CapsDataset(
+    caps_directory=caps_directory,
+    data=sub_ses_t1,
+    preprocessing=preprocessing_t1,
+    transforms=transforms_image,
+)
+dataset_t1_image.prepare_data(n_proc=2)  # to extract the tensor of the T1 file
+
+
+sub_ses_pet_45 = Path(
+    "/Users/camille.brianceau/aramis/CLINICADL/caps/subjects_pet_18FAV45.tsv"
 )
+preprocessing_pet_45 = PreprocessingPET(tracer="18FAV45", suvr_reference_region="pons2")
 
-sub_ses_tsv_1 = Path("")
-split_dir_1 = split_tsv(sub_ses_tsv_1)  # -> creer un test.tsv et un train.tsv
-
-sub_ses_tsv_2 = Path("")
-split_dir_2 = split_tsv(sub_ses_tsv_2)
-
-dataset_t1_roi = caps_reader.get_dataset(
-    preprocessing=preprocessing_1,
-    sub_ses_tsv=split_dir_1 / "train.tsv",
-    transforms=transforms_1,
-)  # do we give config or object for transforms ?
-dataset_pet_patch = caps_reader.get_dataset(
-    preprocessing=preprocessing_2,
-    sub_ses_tsv=split_dir_2 / "train.tsv",
-    transforms=transforms_2,
+dataset_pet_image = CapsDataset(
+    caps_directory=caps_directory,
+    data=sub_ses_pet_45,
+    preprocessing=preprocessing_pet_45,
+    transforms=transforms_image,
 )
+dataset_t1_image.prepare_data(n_proc=2)  # to extract the tensor of the PET file
+
 
-dataset_multi_modality_multi_extract = ConcatDataset(
+dataset_multi_modality = ConcatDataset(
     [
-        dataset_t1_roi,
-        dataset_pet_patch,
-        caps_reader.get_dataset_from_json(json_path=Path("dataset.json")),
+        dataset_t1_image,
+        dataset_pet_image,
     ]
 )  # 3 train.tsv en entrée qu'il faut concat et pareil pour les transforms à faire attention
 
-# TODO : think about adding transforms in extract_json
-
 
 config_file = Path("config_file")
 trainer = Trainer.from_json(config_file=config_file, manager=manager)
@@ -106,6 +90,26 @@
 
 dataloader_config = DataLoaderConfig(n_procs=3, batch_size=10)
 
+
+# CAS 1
+
+# Prérequis : déjà avoir des fichiers avec les listes train et validation
+split_dir = make_kfold(
+    "dataset.tsv"
+)  # lit dataset.tsv => fait le kfold => ecrit la sortie dans split_dir
+splitter = KFolder(
+    dataset_multi_modality, split_dir
+)  # c'est plutôt un iterable de dataloader
+
+# CAS 2
+splitter = KFolder(caps_dataset=dataset_t1_image)
+splitter.make_splits(n_splits=3)
+splitter.write(split_dir)
+
+# or
+splitter = KFolder(caps_dataset=dataset_t1_image)
+splitter.read(split_dir)
+
 for split in splitter.get_splits(splits=(0, 3, 4), dataloader_config=dataloader_config):
     # bien définir ce qu'il y a dans l'objet split
 
@@ -125,19 +129,12 @@
 
 # TEST
 
-preprocessing_test = caps_reader.get_preprocessing("pet-linear")
-transforms_test = Transforms(
-    object_augmentation=[transforms.Crop, transforms.Transform],
-    image_augmentation=[transforms.Crop, transforms.Transform],
-    extraction=ExtractioImageConfig(),
-    image_transforms=[transforms.Blur, transforms.Ghosting],
-    object_transforms=[transforms.BiasField, transforms.Motion],
-)
 
-dataset_test = caps_reader.get_dataset(
-    preprocessing=preprocessing_test,
-    sub_ses_tsv=split_dir_1 / "test.tsv",  # test only on data from the first dataset
-    transforms=transforms_test,
+dataset_test = CapsDataset(
+    caps_directory=caps_directory,
+    preprocessing=preprocessing_t1,
+    sub_ses_tsv=Path("test.tsv"),  # test only on data from the first dataset
+    transforms=transforms_image,
 )
 
 predictor = Predictor(model=model, manager=manager)

diff --git a/clinicadl/API/cross_val.py b/clinicadl/API/cross_val.py
@@ -1,80 +1,38 @@
 from pathlib import Path
 
-import torchio.transforms as transforms
-
-from clinicadl.dataset.caps_dataset import (
-    CapsDatasetPatch,
-    CapsDatasetRoi,
-    CapsDatasetSlice,
-)
-from clinicadl.dataset.caps_reader import CapsReader
-from clinicadl.dataset.concat import ConcatDataset
-from clinicadl.dataset.config.extraction import ExtractionConfig
-from clinicadl.dataset.config.preprocessing import (
-    PreprocessingConfig,
-    T1PreprocessingConfig,
-)
-from clinicadl.dataset.dataloader_config import DataLoaderConfig
+from clinicadl.dataset.datasets.caps_dataset import CapsDataset
 from clinicadl.experiment_manager.experiment_manager import ExperimentManager
-from clinicadl.losses.config import CrossEntropyLossConfig
-from clinicadl.losses.factory import get_loss_function
-from clinicadl.model.clinicadl_model import ClinicaDLModel
-from clinicadl.networks.config import ImplementedNetworks
-from clinicadl.networks.factory import (
-    ConvEncoderOptions,
-    create_network_config,
-    get_network_from_config,
-)
-from clinicadl.optimization.optimizer.config import AdamConfig, OptimizerConfig
-from clinicadl.optimization.optimizer.factory import get_optimizer
 from clinicadl.predictor.predictor import Predictor
-from clinicadl.splitter.kfold import KFolder
-from clinicadl.splitter.split import get_single_split, split_tsv
+from clinicadl.splitter.new_splitter.dataloader import DataLoaderConfig
+from clinicadl.splitter.new_splitter.splitter.kfold import KFold
 from clinicadl.trainer.trainer import Trainer
-from clinicadl.transforms.config import TransformsConfig
 
 # SIMPLE EXPERIMENT WITH A CAPS ALREADY EXISTING
 
 maps_path = Path("/")
 manager = ExperimentManager(maps_path, overwrite=False)
 
-dataset_t1_image = CapsDatasetPatch.from_json(
-    extraction=Path("test.json"),
-    sub_ses_tsv=Path("split_dir") / "train.tsv",
-)
+dataset_t1_image = CapsDataset.from_json(Path("json_path.json"))
+
 config_file = Path("config_file")
 trainer = Trainer.from_json(
     config_file=config_file, manager=manager
 )  # gpu, amp, fsdp, seed
 
-# CAS CROSS-VALIDATION
-splitter = KFolder(caps_dataset=dataset_t1_image, manager=manager)
-split_dir = splitter.make_splits(
-    n_splits=3,
-    output_dir=Path(""),
-    data_tsv=Path("labels.tsv"),
-    subset_name="validation",
-    stratification="",
-)  # Optional data tsv and output_dir
-# n_splits must be >1
-# for the single split case, this method output a path to the directory containing the train and test tsv files so we should have the same output here
+splitter = KFold(dataset=dataset_t1_image)
+splitter.make_splits(n_splits=3)
+split_dir = Path("")
+splitter.write(split_dir)
 
-# CAS EXISTING CROSS-VALIDATION
-splitter = KFolder.from_split_dir(caps_dataset=dataset_t1_image, manager=manager)
+splitter.read(split_dir)
 
 # define the needed parameters for the dataloader
-dataloader_config = DataLoaderConfig(n_procs=3, batch_size=10)
+dataloader_config = DataLoaderConfig(num_workers=3, batch_size=10)
 
-for split in splitter.get_splits(splits=(0, 3, 4), dataloader_config=dataloader_config):
-    # bien définir ce qu'il y a dans l'objet split
 
-    network_config = create_network_config(ImplementedNetworks.CNN)(
-        in_shape=[2, 2, 2],
-        num_outputs=1,
-        conv_args=ConvEncoderOptions(channels=[3, 2, 2]),
-    )
-    optimizer, _ = get_optimizer(network, AdamConfig())
-    model = ClinicaDLModel(network=network_config, loss=nn.MSE(), optimizer=optimizer)
+for split in splitter.get_splits(splits=(0, 3, 4)):
+    print(split)
+    split.build_train_loader(dataloader_config)
+    split.build_val_loader(num_workers=3, batch_size=10)
 
-    trainer.train(model, split)
-    # le trainer va instancier un predictor/valdiator dans le train ou dans le init
+    print(split)
diff --git a/clinicadl/commandline/modules_options/split.py b/clinicadl/commandline/modules_options/split.py
@@ -2,13 +2,14 @@
 
 from clinicadl.config.config_utils import get_default_from_config_class as get_default
 from clinicadl.config.config_utils import get_type_from_config_class as get_type
-from clinicadl.splitter.config import SplitConfig
+from clinicadl.splitter.splitter.kfold import KFoldConfig
+from clinicadl.splitter.splitter.single_split import SingleSplitConfig
 
 # Cross Validation
 n_splits = click.option(
     "--n_splits",
-    type=get_type("n_splits", SplitConfig),
-    default=get_default("n_splits", SplitConfig),
+    type=get_type("n_splits", KFoldConfig),
+    default=get_default("n_splits", KFoldConfig),
     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,
@@ -17,7 +18,7 @@
     "--split",
     "-s",
     type=int,  # get_type("split", config.ValidationConfig),
-    default=get_default("split", SplitConfig),
+    default=get_default("split", SingleSplitConfig),
     multiple=True,
     help="Train the list of given splits. By default, all the splits are trained.",
     show_default=True,

diff --git a/clinicadl/dataset/dataloader/__init__.py b/clinicadl/dataset/dataloader/__init__.py
@@ -0,0 +1 @@
+from .config import DataLoaderConfig