fixing moshiko PR review

fuse_examples/imaging/segmentation/picai/conf/config_template.yaml

@@ -24,12 +24,12 @@ train:
   learning_rate : 1e-3
   weight_decay :  0
   run_sample : 0 #if 0 - used all samples otherwise sample only run_sample samples ( for test purpouse )
-  resume_checkpoint_filename : 
+  resume_checkpoint_filename : null
   trainer:
     accelerator : gpu
     devices : 1
     num_epochs : 100
-    ckpt_path : 
+    ckpt_path : null
   unet_kwargs :
     strides : [[2, 2, 2], [1, 2, 2], [1, 2, 2], [1, 2, 2], [2, 2, 2]]
     channels : [32, 64, 128, 256, 512, 1024]

fuse_examples/imaging/segmentation/picai/runner.py

@@ -144,7 +144,7 @@ def run_train(paths: NDict, train: NDict) -> None:
     lgr.info("- Create sampler:")
     sampler = BatchSamplerDefault(
         dataset=train_dataset,
-        balanced_class_name="data.gt.classification",  # gt_label, TODO - make diff label for balance-sampler
+        balanced_class_name="data.gt.classification",
         num_balanced_classes=2,
         batch_size=train["batch_size"],
         mode="approx",
@@ -199,7 +199,7 @@ def run_train(paths: NDict, train: NDict) -> None:
 
     # either a dict with arguments to pass to ModelCheckpoint or list dicts for multiple ModelCheckpoint callbacks (to monitor and save checkpoints for more then one metric).
     best_epoch_source = dict(
-        monitor="validation.losses.total_loss",  # metrics.auc.macro_avg",
+        monitor="validation.losses.total_loss",
         mode="min",
     )
 
@@ -331,7 +331,6 @@ def run_eval(paths: NDict, infer: NDict) -> None:
     # define iterator
 
     def data_iter() -> NDict:
-        # set seed
         data_file = os.path.join(paths["inference_dir"], "infer.pickle")
         data = pd.read_pickle(data_file)
         for fold in data:
@@ -372,7 +371,7 @@ def main(cfg: DictConfig) -> None:
     if "infer" in cfg["run.running_modes"]:
         run_infer(NDict(cfg["infer"]), NDict(cfg["paths"]), NDict(cfg["train"]))
 
-    # analyze - skipping as it crushes without metrics
+    # analyze
     if "eval" in cfg["run.running_modes"]:
         run_eval(NDict(cfg["paths"]), NDict(cfg["infer"]))
 

fuseimg/datasets/picai.py

@@ -298,13 +298,12 @@ def dataset(
     ) -> DatasetDefault:
         """
         Creates Fuse Dataset single object (either for training, validation and test or user defined set)
-        :param data_dir:                    dataset root path
-        :param clinical_file                path to clinical_file
-        :param target                       target name used from the ground truth dataframe
-        :param cache_dir:                   Optional, name of the cache folder
+        :param paths                        paths dictionary for dataset files
+        :param cfg                          dict cfg for training phase
         :param reset_cache:                 Optional,specifies if we want to clear the cache first
-        :param sample_ids: dataset including the specified sample_ids or None for all the samples. sample_id is case_{id:05d} (for example case_00001 or case_00100).
-        :param train: True if used for training  - adds augmentation operations to the pipeline
+        :param sample_ids:                  dataset including the specified sample_ids or None for all the samples.
+        :param train:                       True if used for training  - adds augmentation operations to the pipeline
+        :param run_sample:                  if > 0 it samples from all the samples #run_sample examples ( used for testing), if =0 then it takes all samples
         :return: DatasetDefault object
         """