normalize features before clustering

src/depiction/tools/clustering.py

@@ -12,6 +12,7 @@
 from depiction.clustering.metrics import cross_correlation
 from depiction.clustering.stratified_grid import StratifiedGrid
 from depiction.image.feature_selection import FeatureSelectionIQR, retain_features
+from depiction.image.image_normalization import ImageNormalization, ImageNormalizationVariant
 from depiction.image.multi_channel_image import MultiChannelImage
 
 
@@ -62,6 +63,9 @@ def clustering(
     image_full_combined = MultiChannelImage.read_hdf5(path=input_hdf5)
     assert "cluster" not in image_full_combined.channel_names
     image_full_features = image_full_combined.drop_channels(coords=["image_index"], allow_missing=True)
+    image_full_features = ImageNormalization().normalize_image(
+        image=image_full_features, variant=ImageNormalizationVariant.VEC_NORM
+    )
     image_full_image_index = image_full_combined.retain_channels(coords=["image_index"])
 
     # retain the most relevant features
@@ -100,18 +104,11 @@ def clustering(
 
     # write the result of the operation
     output_image = MultiChannelImage(
-        xarray.concat([label_image.data_spatial, image_full_features.data_spatial], dim="c")
+        xarray.concat([label_image.data_spatial, image_full_combined.data_spatial], dim="c")
     )
     output_image.write_hdf5(output_hdf5)
 
 
-# def select_features_cv(image: MultiChannelImage, n_keep: int = 30) -> DataArray:
-#    image_data = image.data_flat
-#    cv_score = image_data.std("i") / image_data.mean("i")
-#    return image.retain_channels(coords=cv_score.sortby("c")[-n_keep:].c.values)
-#
-
-
 def compute_labels(features: NDArray[float], method: MethodEnum, method_params: str) -> NDArray[int]:
     if method == MethodEnum.KMEANS:
         clu = KMeans(n_clusters=10).fit(features)