dataeval 0.86.5__py3-none-any.whl → 0.86.7__py3-none-any.whl

dataeval/__init__.py

@@ -8,7 +8,7 @@ shifts that impact performance of deployed models.
 from __future__ import annotations
 
 __all__ = ["config", "detectors", "log", "metrics", "typing", "utils", "workflows"]
-__version__ = "0.86.5"
+__version__ = "0.86.7"
 
 import logging
 

dataeval/data/_metadata.py

@@ -234,14 +234,17 @@ class Metadata:
             if is_od_target := isinstance(target, ObjectDetectionTarget):
                 target_labels = as_numpy(target.labels)
                 target_len = len(target_labels)
-                labels.extend(target_labels.tolist())
-                bboxes.extend(as_numpy(target.boxes).tolist())
-                scores.extend(as_numpy(target.scores).tolist())
-                srcidx.extend([i] * target_len)
+                if target_len:
+                    labels.extend(target_labels.tolist())
+                    bboxes.extend(as_numpy(target.boxes).tolist())
+                    scores.extend(as_numpy(target.scores).tolist())
+                    srcidx.extend([i] * target_len)
             elif isinstance(target, Array):
-                target_len = 1
-                labels.append(int(np.argmax(as_numpy(target))))
-                scores.append(target)
+                if len(target):
+                    target_len = 1
+                    labels.append(int(np.argmax(as_numpy(target))))
+                    scores.append(target)
+                    srcidx.append(i)
             else:
                 raise TypeError("Encountered unsupported target type in dataset")
 
@@ -252,18 +255,18 @@ class Metadata:
         labels = as_numpy(labels).astype(np.intp)
         scores = as_numpy(scores).astype(np.float32)
         bboxes = as_numpy(bboxes).astype(np.float32) if is_od else None
-        srcidx = as_numpy(srcidx).astype(np.intp) if is_od else None
+        srcidx = as_numpy(srcidx).astype(np.intp)
 
         index2label = self._dataset.metadata.get("index2label", {i: str(i) for i in np.unique(labels)})
 
-        targets_per_image = None if srcidx is None else np.unique(srcidx, return_counts=True)[1].tolist()
+        targets_per_image = np.bincount(srcidx, minlength=len(self._dataset)).tolist() if is_od else None
         merged = merge(raw, return_dropped=True, ignore_lists=False, targets_per_image=targets_per_image)
 
         reserved = ["image_index", "class_label", "score", "box"]
         factor_dict = {f"metadata_{k}" if k in reserved else k: v for k, v in merged[0].items() if k != "_image_index"}
 
         target_dict = {
-            "image_index": srcidx if srcidx is not None else np.arange(len(labels)),
+            "image_index": srcidx,
             "class_label": labels,
             "score": scores,
             "box": bboxes if bboxes is not None else [None] * len(labels),

dataeval/detectors/linters/outliers.py

@@ -18,57 +18,59 @@ from dataeval.outputs._stats import BASE_ATTRS
 from dataeval.typing import ArrayLike, Dataset
 
 
+def _get_zscore_mask(values: NDArray[np.float64], threshold: float | None) -> NDArray[np.bool_] | None:
+    threshold = threshold if threshold is not None else 3.0
+    std_val = np.nanstd(values)
+    if std_val > EPSILON:
+        mean_val = np.nanmean(values)
+        abs_diff = np.abs(values - mean_val)
+        return (abs_diff / std_val) > threshold
+    return None
+
+
+def _get_modzscore_mask(values: NDArray[np.float64], threshold: float | None) -> NDArray[np.bool_] | None:
+    threshold = threshold if threshold is not None else 3.5
+    median_val = np.nanmedian(values)
+    abs_diff = np.abs(values - median_val)
+    m_abs_diff = np.nanmedian(abs_diff)
+    m_abs_diff = np.nanmean(abs_diff) if m_abs_diff <= EPSILON else m_abs_diff
+    if m_abs_diff > EPSILON:
+        mod_z_score = 0.6745 * abs_diff / m_abs_diff
+        return mod_z_score > threshold
+    return None
+
+
+def _get_iqr_mask(values: NDArray[np.float64], threshold: float | None) -> NDArray[np.bool_] | None:
+    threshold = threshold if threshold is not None else 1.5
+    qrt = np.nanpercentile(values, q=(25, 75), method="midpoint")
+    iqr_val = qrt[1] - qrt[0]
+    if iqr_val > EPSILON:
+        iqr_threshold = iqr_val * threshold
+        return (values < (qrt[0] - iqr_threshold)) | (values > (qrt[1] + iqr_threshold))
+    return None
+
+
 def _get_outlier_mask(
     values: NDArray[Any], method: Literal["zscore", "modzscore", "iqr"], threshold: float | None
 ) -> NDArray[np.bool_]:
     if len(values) == 0:
         return np.array([], dtype=bool)
 
-    values = values.astype(np.float64)
-
-    valid_mask = ~np.isnan(values)
-    outliers = np.full(values.shape, False, dtype=bool)
-
-    if not np.any(valid_mask):
-        return outliers
-
-    if method == "zscore":
-        threshold = threshold if threshold is not None else 3.0
-
-        std_val = np.nanstd(values)
-
-        if std_val > EPSILON:
-            mean_val = np.nanmean(values)
-            abs_diff = np.abs(values - mean_val)
-            outliers = (abs_diff / std_val) > threshold
+    nan_mask = np.isnan(values)
 
+    if np.all(nan_mask):
+        outliers = None
+    elif method == "zscore":
+        outliers = _get_zscore_mask(values.astype(np.float64), threshold)
     elif method == "modzscore":
-        threshold = threshold if threshold is not None else 3.5
-
-        median_val = np.nanmedian(values)
-        abs_diff = np.abs(values - median_val)
-        m_abs_diff = np.nanmedian(abs_diff)
-        m_abs_diff = np.nanmean(abs_diff) if m_abs_diff <= EPSILON else m_abs_diff
-
-        if m_abs_diff > EPSILON:
-            mod_z_score = 0.6745 * abs_diff / m_abs_diff
-            outliers = mod_z_score > threshold
-
+        outliers = _get_modzscore_mask(values.astype(np.float64), threshold)
     elif method == "iqr":
-        threshold = threshold if threshold is not None else 1.5
-
-        qrt = np.nanpercentile(values, q=(25, 75), method="midpoint")
-        iqr_val = qrt[1] - qrt[0]
-
-        if iqr_val > EPSILON:
-            iqr_threshold = iqr_val * threshold
-            outliers = (values < (qrt[0] - iqr_threshold)) | (values > (qrt[1] + iqr_threshold))
-
+        outliers = _get_iqr_mask(values.astype(np.float64), threshold)
     else:
         raise ValueError("Outlier method must be 'zscore' 'modzscore' or 'iqr'.")
 
-    outliers[~valid_mask] = False
-    return outliers
+    # If outliers were found, return the mask with NaN values set to False, otherwise return all False
+    return outliers & ~nan_mask if outliers is not None else np.full(values.shape, False, dtype=bool)
 
 
 class Outliers:

dataeval/metrics/stats/_dimensionstats.py

@@ -6,7 +6,6 @@ from typing import Any, Callable
 
 import numpy as np
 
-from dataeval.config import EPSILON
 from dataeval.metrics.stats._base import StatsProcessor, run_stats
 from dataeval.outputs import DimensionStatsOutput
 from dataeval.outputs._base import set_metadata
@@ -23,8 +22,8 @@ class DimensionStatsProcessor(StatsProcessor[DimensionStatsOutput]):
         "height": lambda x: x.box.height,
         "channels": lambda x: x.shape[-3],
         "size": lambda x: x.box.width * x.box.height,
-        "aspect_ratio": lambda x: x.box.width / (x.box.height + EPSILON),
-        "depth": lambda x: get_bitdepth(x.image).depth,
+        "aspect_ratio": lambda x: 0.0 if x.box.height == 0 else x.box.width / x.box.height,
+        "depth": lambda x: get_bitdepth(x.raw).depth,
         "center": lambda x: np.asarray([(x.box.x0 + x.box.x1) / 2, (x.box.y0 + x.box.y1) / 2]),
         "distance_center": lambda x: np.sqrt(
             np.square(((x.box.x0 + x.box.x1) / 2) - (x.raw.shape[-1] / 2))

{dataeval-0.86.5.dist-info → dataeval-0.86.7.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: dataeval
-Version: 0.86.5
+Version: 0.86.7
 Summary: DataEval provides a simple interface to characterize image data and its impact on model performance across classification and object-detection tasks
 Home-page: https://dataeval.ai/
 License: MIT

{dataeval-0.86.5.dist-info → dataeval-0.86.7.dist-info}/RECORD

@@ -1,10 +1,10 @@
-dataeval/__init__.py,sha256=5qOVdEDEZt5O--VufuRJXGEByzQC7pJWZluFGzPuNOc,1636
+dataeval/__init__.py,sha256=P6WvVjHlE2nH57bXBR4A9ez6R32OQGm9bshYrxRKwFw,1636
 dataeval/_log.py,sha256=C7AGkIRzymvYJ0LQXtnShiy3i5Xrp8T58JzIHHguk_Q,365
 dataeval/config.py,sha256=bHa8np4FCtLLv8_xlfdDC4lb1InJ_kT0vXDO5P42rvk,4082
 dataeval/data/__init__.py,sha256=wzQ6uUFLNB3VJR0a2QnRBYwEmwXT93q0WpHu7FmFW1E,486
 dataeval/data/_embeddings.py,sha256=PFjpdV9bfusCB4taTIYSzx1hP8nJb_KCkZTN8kMw-Hs,12885
 dataeval/data/_images.py,sha256=Rc_59CuU4zfN7Xm7an1XUx8ZghQg6a56VJWMZD9edRw,2654
-dataeval/data/_metadata.py,sha256=OTda9V7DA5Ejxip_NR16LCK2C8HMtpjWHHiFoW3LrLY,14364
+dataeval/data/_metadata.py,sha256=3aixstlgcAZXC0qNjwDlxjscC3IX1xjPt_FK0liRqoo,14423
 dataeval/data/_selection.py,sha256=r06xeiyK8nTWPLyItkoPQRWZI1i6LATSue_cuEbCdc4,4463
 dataeval/data/_split.py,sha256=nQABR05vxil2Qx7-uX4Fm0_DWpibskBGDJOYj_b1u3I,16737
 dataeval/data/selections/__init__.py,sha256=2m8ZB53wXzqLcqmc6p5atO6graB6ZyiRSNJFxf11X_g,613
@@ -32,7 +32,7 @@ dataeval/detectors/drift/_uncertainty.py,sha256=BHlykJ-r7TGLJxdPfoazXnoAJ1qVDzbk
 dataeval/detectors/drift/updates.py,sha256=L1PnrPlIE1x6ujCc5mCwjcAZwadVTn-Zjb6MnTDvzJQ,2251
 dataeval/detectors/linters/__init__.py,sha256=xn2zPwUcmsuf-Jd9uw6AVI11C9z1b1Y9fYtuFnXenZ0,404
 dataeval/detectors/linters/duplicates.py,sha256=X5WSEvI_BHkLoXjkaHK6wTnSkx4IjpO_exMRjSlhc70,4963
-dataeval/detectors/linters/outliers.py,sha256=WO686jVbGbtDjO-8CuYVLxpeUGv8MpIK9QjADlTdd40,9596
+dataeval/detectors/linters/outliers.py,sha256=GaM9n8yPgBPzVOL_bxJCj0eCwobEEP4JHKHD9liRdlw,10130
 dataeval/detectors/ood/__init__.py,sha256=juCYBDs7CQEAtMhnEpPqF6uTrOIH9kTBSuQ_GRw6a8o,283
 dataeval/detectors/ood/ae.py,sha256=fTrUfFxv6xUqzKpwMC8rW3JrizA16M_bgzqLuBKMrS0,2944
 dataeval/detectors/ood/base.py,sha256=9b-Ljznf0lB1SXF4F_Aj3eJ4Y3ijGEDPMjucUsWOGJM,3051
@@ -56,7 +56,7 @@ dataeval/metrics/estimators/_uap.py,sha256=BULEBbJ9BQ1IcTeZf0x7iI60QHAWCccBOM97F
 dataeval/metrics/stats/__init__.py,sha256=6tA_9nbbM5ObJ6cds8Y1VBtTQiTOxrpGQSFLu_lWGGA,1098
 dataeval/metrics/stats/_base.py,sha256=R-hxoEPLreZcxYxBfyjbKfdoGMMTPiqJ5g2zSO-1UYM,12541
 dataeval/metrics/stats/_boxratiostats.py,sha256=ROZrlqgbowkGfCR5PJ5TL7Og40iMOdUqJnsCtaz_Xek,6450
-dataeval/metrics/stats/_dimensionstats.py,sha256=EVO-BlxrZl8qrP09lwPbyWdrG1ZeDtgj4LiswDwEZ1I,2896
+dataeval/metrics/stats/_dimensionstats.py,sha256=s2Juca8GG501nZd2SWL_YtXWkTfxUrUIAl53PO3_VeA,2876
 dataeval/metrics/stats/_hashstats.py,sha256=qa1CYRgOebkxqkALfffaPM-kJ074ZbyfpWbfOfuObSs,4758
 dataeval/metrics/stats/_imagestats.py,sha256=gUPNgN5Zwzdr7WnSwbve1NXNsyxd5dy3cSnlR_7guCg,3007
 dataeval/metrics/stats/_labelstats.py,sha256=_dXt3p8_-SHEtHvJWbL0rnQvO2g30zxX42mG2LGJepU,3195
@@ -107,7 +107,7 @@ dataeval/utils/torch/models.py,sha256=1idpXyjrYcCBSsbxxRUOto8xr4MJNjDEqQHiIXVU5Z
 dataeval/utils/torch/trainer.py,sha256=Oc2lK13uPGhmLYbmAqlPWyKxgG4YJFlnSXCqFHUZbdA,5528
 dataeval/workflows/__init__.py,sha256=ou8y0KO-d6W5lgmcyLjKlf-J_ckP3vilW7wHkgiDlZ4,255
 dataeval/workflows/sufficiency.py,sha256=j-R8dg4XE6a66p_oTXG2GNzgg3vGk85CTblxhFXaxog,8513
-dataeval-0.86.5.dist-info/LICENSE.txt,sha256=uAooygKWvX6NbU9Ran9oG2msttoG8aeTeHSTe5JeCnY,1061
-dataeval-0.86.5.dist-info/METADATA,sha256=qx7aNDgzyAfRRKWjDXkfXojBdsBFnjMgwTVl0JsLbbw,5353
-dataeval-0.86.5.dist-info/WHEEL,sha256=Nq82e9rUAnEjt98J6MlVmMCZb-t9cYE2Ir1kpBmnWfs,88
-dataeval-0.86.5.dist-info/RECORD,,
+dataeval-0.86.7.dist-info/LICENSE.txt,sha256=uAooygKWvX6NbU9Ran9oG2msttoG8aeTeHSTe5JeCnY,1061
+dataeval-0.86.7.dist-info/METADATA,sha256=7FTgPB4Yj2zF7z2B6IIRe9WFc9VCBqrcFEIf5ByVHdw,5353
+dataeval-0.86.7.dist-info/WHEEL,sha256=Nq82e9rUAnEjt98J6MlVmMCZb-t9cYE2Ir1kpBmnWfs,88
+dataeval-0.86.7.dist-info/RECORD,,