dataeval 1.0.1__tar.gz → 1.0.2__tar.gz

{dataeval-1.0.1 → dataeval-1.0.2}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: dataeval
-Version: 1.0.1
+Version: 1.0.2
 Summary: DataEval provides a simple interface to characterize image data and its impact on model performance across classification and object-detection tasks
 Project-URL: Homepage, https://dataeval.ai/
 Project-URL: Repository, https://github.com/aria-ml/dataeval/

{dataeval-1.0.1 → dataeval-1.0.2}/pyproject.toml

@@ -103,7 +103,7 @@ docs = [
   "jinja2>=3.1.6",
   "jupyter-client>=8.6.0",
   "jupyter-cache>=1.0",
-  "maite-datasets>=0.0.10",
+  "maite-datasets>=0.0.12",
   "myst-nb>=1.0",
   "opencv-python-headless>=4.8.0",
   "plotly>=6.2.0",

{dataeval-1.0.1 → dataeval-1.0.2}/src/dataeval/_metadata.py

@@ -1311,7 +1311,13 @@ class Metadata(Array, FeatureExtractor):
                 n_classes,
             )
 
-        index2label = self._dataset.metadata.get("index2label", {i: str(i) for i in np.unique(labels)})
+        unique_labels = np.unique(labels) if len(labels) else np.array([], dtype=np.intp)
+        provided_i2l = self._dataset.metadata.get("index2label", None)
+        if provided_i2l is not None:
+            # Ensure every observed label has a name; use fallback for unmapped labels
+            index2label = {int(lbl): provided_i2l.get(int(lbl), f"UNDEFINED_CLASS_{int(lbl)}") for lbl in unique_labels}
+        else:
+            index2label = {int(lbl): str(int(lbl)) for lbl in unique_labels}
         target_idx = self._compute_target_indices(srcidx, datum_count, bool(self._has_targets))
         reserved = ["item_index", "target_index", "class_label", "score", "box"]
         target_factor_dict = {}

{dataeval-1.0.1 → dataeval-1.0.2}/src/dataeval/_version.py

@@ -28,7 +28,7 @@ version_tuple: VERSION_TUPLE
 commit_id: COMMIT_ID
 __commit_id__: COMMIT_ID
 
-__version__ = version = '1.0.1'
-__version_tuple__ = version_tuple = (1, 0, 1)
+__version__ = version = '1.0.2'
+__version_tuple__ = version_tuple = (1, 0, 2)
 
 __commit_id__ = commit_id = None

{dataeval-1.0.1 → dataeval-1.0.2}/src/dataeval/core/_compute_stats.py

@@ -351,6 +351,11 @@ def compute_stats(
     """
     Compute specified statistics on a set of images, optionally within bounding boxes.
 
+    Mixed-bit-depth datasets can produce misleading statistics when raw pixel values are
+    compared directly. To avoid this, pixel values are normalized to [0, 1] based on each
+    image's bit depth before any statistic is computed, keeping results meaningful and
+    comparable across 8-bit, 16-bit, 32-bit, and other precision images.
+
     Parameters
     ----------
     data : Iterable[ArrayLike] | Dataset[ArrayLike] | Dataset[tuple[ArrayLike, Any, Any]]

{dataeval-1.0.1 → dataeval-1.0.2}/src/dataeval/core/_parity.py

@@ -116,6 +116,7 @@ def parity(
     chi_scores = np.zeros(factor_data_np.shape[1])
     p_values = np.zeros_like(chi_scores)
     insufficient_ddict: defaultdict[int, defaultdict[int, dict[int, int]]] = defaultdict(lambda: defaultdict(dict))
+    unique_class_labels = np.unique(class_labels_np)
 
     for i, col_data in enumerate(factor_data_np.T):
         # Builds a contingency matrix where entry at index (r,c) represents
@@ -132,7 +133,8 @@ def parity(
             if contingency_matrix[int_factor, int_class] > 0:
                 factor_category = unique_factor_values[int_factor].item()
                 class_count = contingency_matrix[int_factor, int_class].item()
-                insufficient_ddict[i][factor_category][int_class] = class_count
+                class_label = int(unique_class_labels[int_class])
+                insufficient_ddict[i][factor_category][class_label] = class_count
 
         # This deletes rows containing only zeros,
         # because scipy.stats.chi2_contingency fails when there are rows containing only zeros.