Add CLI for five-fold VS training

research/vestibular_schwannoma/README.md

@@ -6,11 +6,12 @@ training, inference on new cases, and PACS deployment.
 
 ## Contents
 
+- `train_5fold.py`: command-line five-fold training and evaluation.
 - `notebooks/01_five_fold_cross_validation.ipynb`: train and compare UNet, DynUNet, and
   optional SegMamba models.
 - `notebooks/02_inference_new_cases.ipynb`: run one declared model or an explicit ensemble.
 - `workflow/`: project-local configuration, training model definitions, result aggregation,
-  and inference artifact handling used by the notebooks.
+  and inference artifact handling shared by the CLI and notebooks.
 - `data/ml_dataset.csv`: public case index and fixed fold assignments.
 - `deployment/pacs/`: Safetensors bundle builder and ROR/PACS container.
 - `tests/workflow/`: CPU-only workflow contract and orchestration tests.
@@ -32,7 +33,18 @@ optional SegMamba fork when needed.
 
 ## Run
 
-Start Jupyter from this directory or `notebooks/`:
+Use the CLI for unattended training. Examples for a quick check, one complete model, and the
+full comparison:
+
+```bash
+python train_5fold.py --models unet --folds 1 --epochs 5 --no-compile
+python train_5fold.py --models unet  # One model, all five folds
+python train_5fold.py --skip-unavailable
+```
+
+The default requests four models across five folds for 500 epochs; run
+`python train_5fold.py --help` before starting. For interactive inspection and visualizations,
+start Jupyter from this directory or `notebooks/`:
 
 ```bash
 jupyter lab notebooks/01_five_fold_cross_validation.ipynb

research/vestibular_schwannoma/tests/workflow/test_config.py

@@ -22,6 +22,7 @@ def test_defaults_preserve_the_notebook_experiment(self):
         self.assertEqual(config.training_seed, 42)
         self.assertEqual(config.queue_num_workers, 4)
         self.assertEqual(config.queue_length, 300)
+        self.assertEqual(config.foreground_sampling_probability, 0.8)
         self.assertTrue(config.use_tta)
 
     def test_configuration_is_frozen(self):
@@ -44,6 +45,14 @@ def test_invalid_declarations_fail_early(self):
             {"model_keys": ("unet",), "training_seed": -1},
             {"model_keys": ("unet",), "training_seed": True},
             {"model_keys": ("unet",), "patch_size": (192, 192, 0)},
+            {
+                "model_keys": ("unet",),
+                "foreground_sampling_probability": 0,
+            },
+            {
+                "model_keys": ("unet",),
+                "foreground_sampling_probability": 1,
+            },
         ]
         for kwargs in cases:
             with self.subTest(kwargs=kwargs), self.assertRaises(ValueError):
@@ -71,6 +80,14 @@ def test_patch_factory_preserves_training_and_inference_contract(self):
             },
         )
 
+    def test_patch_factory_supports_70_30_sampling(self):
+        config = ExperimentConfig(
+            model_keys=("unet",), foreground_sampling_probability=0.7
+        )
+        patch = make_patch_config(config, [])
+        self.assertAlmostEqual(patch.label_probabilities[0], 0.3)
+        self.assertAlmostEqual(patch.label_probabilities[1], 0.7)
+
 
 if __name__ == "__main__":
     unittest.main()

research/vestibular_schwannoma/tests/workflow/test_models.py

@@ -8,10 +8,15 @@ class TrainingModelConfigTests(unittest.TestCase):
     def test_specs_are_the_single_architecture_declaration(self):
         self.assertEqual(models.UNET_SPEC["arch_id"], "monai.unet")
         self.assertEqual(models.DYNUNET_SPEC["arch_id"], "monai.dynunet")
+        self.assertEqual(models.DYNUNET_SMALL_SPEC["arch_id"], "monai.dynunet")
         self.assertEqual(
             models.DYNUNET_SPEC["wrapper_spec"][0]["wrapper_id"],
             "fastmonai.dynunet_ds_adapter",
         )
+        self.assertEqual(
+            models.DYNUNET_SMALL_SPEC["arch_kwargs"]["filters"],
+            [32, 64, 128, 256, 512],
+        )
         self.assertEqual(models.SEGMAMBA_SPEC["arch_id"], "segmamba.v2")
 
     def test_declared_order_is_preserved(self):

research/vestibular_schwannoma/tests/workflow/test_train_5fold.py

@@ -0,0 +1,27 @@
+import unittest
+
+from vestibular_schwannoma import train_5fold
+
+
+class FiveFoldLauncherTests(unittest.TestCase):
+    def test_defaults_select_all_models_and_preserve_80_20_control(self):
+        args = train_5fold._parser().parse_args([])
+
+        self.assertEqual(tuple(args.models), train_5fold.DEFAULT_MODELS)
+        self.assertEqual(args.folds, [1, 2, 3, 4, 5])
+        self.assertEqual(args.foreground_probability, 0.8)
+        self.assertNotIn("use_tta", vars(args))
+
+    def test_70_30_sampling_remains_an_explicit_comparison(self):
+        args = train_5fold._parser().parse_args(["--foreground-probability", "0.7"])
+
+        self.assertEqual(args.foreground_probability, 0.7)
+
+    def test_small_dynunet_uses_the_intermediate_512_bottleneck(self):
+        spec = train_5fold.TRAINING_MODEL_CONFIGS["dynunet_small"].model_spec
+
+        self.assertEqual(spec["arch_kwargs"]["filters"], [32, 64, 128, 256, 512])
+
+
+if __name__ == "__main__":
+    unittest.main()

research/vestibular_schwannoma/tests/workflow/test_training.py

@@ -232,6 +232,7 @@ def test_fold_tracking_uses_fixed_model_and_output_contract(self):
         )
         extra_params = create_callback.call_args.kwargs["extra_params"]
         self.assertEqual(extra_params["training_seed"], 42)
+        self.assertEqual(extra_params["foreground_sampling_probability"], 0.8)
         self.assertEqual(
             json.loads(extra_params["gpu_augmentations"]),
             [
@@ -332,6 +333,7 @@ def build_model(*args, **kwargs):
         )
         extra_params = create_callback.call_args.kwargs["extra_params"]
         self.assertEqual(extra_params["training_seed"], 42)
+        self.assertEqual(extra_params["foreground_sampling_probability"], 0.8)
         self.assertEqual(
             json.loads(extra_params["gpu_augmentations"]),
             [