Merge branch 'main' into linear_probe_exposure

Author: Sllambias

.github/workflows/ci.yaml

@@ -35,6 +35,3 @@ jobs:
         with:
           args: format --check .
 
-      - name: Run tests
-        if: ${{ always() }} 
-        run: pytest

.github/workflows/pipeline_tests.yml

@@ -0,0 +1,51 @@
+name: Pipeline tests
+
+on:
+  push:
+    branches: [main]
+  pull_request:
+    branches: [main]
+
+env:
+  ASPARAGUS_MODELS: "${{ github.workspace }}/models"
+  #ASPARAGUS: "${{ github.workspace }}/tests/models"
+  HYDRA_FULL_ERROR: 1
+
+jobs:
+  pipeline-tests:
+    runs-on: ubuntu-latest
+    timeout-minutes: 20
+
+    steps:
+      - uses: actions/checkout@v4
+
+      - uses: actions/setup-python@v5
+        with:
+          python-version: "3.11"
+          cache: pip
+
+      - name: Install CPU-only torch then package
+        run: |
+          pip install torch torchvision --index-url https://download.pytorch.org/whl/cpu
+          pip install -e ".[test]"
+
+      - name: Run pretrain test
+        run: |
+          pytest tests/test_pretrain.py -v --timeout=300
+
+      - name: Run finetune_seg test
+        run: |
+          pytest tests/test_finetune_seg.py -v --timeout=300
+
+      - name: Run train_reg test
+        run: |
+          pytest tests/test_train_reg.py -v --timeout=300
+
+      - name: Run test_cls test
+        run: |
+          pytest tests/test_test_cls.py -v --timeout=300
+
+      - name: Run linear probe test
+        run: |
+          pytest tests/test_linear_probe.py -v --timeout=300
+

.github/workflows/stale.yaml

@@ -8,15 +8,15 @@ def reverse_preprocessing(array, image_properties):
     pad_bbox = image_properties["pad_box"]
     crop_bbox = image_properties["crop_box"]
 
-    shape = array.shape[2:]
-    if len(shape) == 2:
+    ndim = len(array.shape[2:])
+    if ndim == 2:
         mode = "bilinear"
-    elif len(shape) == 3:
+    elif ndim == 3:
         mode = "trilinear"
 
     if len(pad_bbox) > 0:
         array = unpad_array(array, pad_bbox)
-        verify_shapes_are_equal(reference_shape=shape, target_shape=image_properties["shape_before_pad"])
+        verify_shapes_are_equal(reference_shape=array.shape[2:], target_shape=image_properties["shape_before_pad"])
 
     array = F.interpolate(array, size=image_properties["size_before_resample"], mode=mode)
 

asparagus/functional/reverse_preprocessing.py

@@ -11,6 +11,7 @@
 from torchmetrics.classification import (
     MulticlassAUROC,
     MulticlassAveragePrecision,
+    MulticlassF1Score,
 )
 from torchvision import transforms
 from typing import List, Optional
@@ -198,6 +199,7 @@ def configure_test_metrics(self):
             {
                 "AUROC_macro": MulticlassAUROC(num_classes=self.num_classes, average="macro"),
                 "AUPRC_macro": MulticlassAveragePrecision(num_classes=self.num_classes, average="macro"),
+                "F1_macro": MulticlassF1Score(num_classes=self.num_classes, average="macro"),
             }
         )
 
@@ -211,6 +213,7 @@ def configure_metrics(self, prefix: str):
                     f"{prefix}/{head_name}/auprc_macro": MulticlassAveragePrecision(
                         num_classes=self.num_classes, average="macro"
                     ),
+                    f"{prefix}/{head_name}/f1_macro": MulticlassF1Score(num_classes=self.num_classes, average="macro"),
                 }
             )
         return metrics

asparagus/modules/lightning_modules/linear_probe_module.py

@@ -23,9 +23,6 @@
     volume_similarity,
 )
 from gardening_tools.functional.paths.write import save_json
-from gardening_tools.functional.transforms.cropping_and_padding import (
-    fit_patch_size_to_image_size,
-)
 from gardening_tools.modules.losses.deep_supervision import DeepSupervisionLoss
 from gardening_tools.modules.losses.DiceCE import DiceCE
 from gardening_tools.modules.metrics import GeneralizedDiceScore
@@ -51,7 +48,6 @@ def __init__(
         val_transforms: Optional[transforms.Compose] = None,
         optimizer: str = "SGD",
         inference_patch_size: list = [],
-        inference_mode: str = "3D",
         test_output_path: str = None,
         log_image_every_n_epochs: int = 50,
         weight_decay: float = 3e-5,
@@ -78,7 +74,6 @@ def __init__(
             load_decoder=load_decoder,
             repeat_stem_weights=repeat_stem_weights,
         )
-        self.inference_mode = inference_mode
         self.inference_patch_size = inference_patch_size
         self.test_output_path = test_output_path
         self.num_classes = model.num_classes
@@ -161,7 +156,6 @@ def training_step(self, batch, batch_idx):
 
     def validation_step(self, batch, batch_idx):
         x, y = batch["image"], batch["label"]
-
         pred = self.model(x)
         loss = self.val_loss(pred, y)
         self.log(
@@ -221,7 +215,7 @@ def test_step(self, batch, batch_idx):
 
         logits = self.model.sliding_window_predict(
             data=x,
-            patch_size=fit_patch_size_to_image_size(self.inference_patch_size, list(x.shape[2:])),
+            patch_size=self.inference_patch_size,
             overlap=0.5,
         )
 

asparagus/modules/lightning_modules/segmentation_module.py

@@ -1 +1,3 @@
 from asparagus.modules.transforms.clamp import Torch_ClampTarget as Torch_ClampTarget
+from asparagus.modules.transforms.crop import Torch_Crop as Torch_Crop
+from asparagus.modules.transforms.pad import Torch_Pad as Torch_Pad