valor-lite 0.33.11__py3-none-any.whl → 0.33.13__py3-none-any.whl

valor_lite/classification/computation.py

@@ -182,9 +182,9 @@ def compute_metrics(
         out=precision,
     )
 
-    accuracy = np.zeros_like(recall)
+    accuracy = np.zeros(n_scores, dtype=np.float64)
     np.divide(
-        (counts[:, :, 0] + counts[:, :, 3]),
+        counts[:, :, 0].sum(axis=1),
         float(n_datums),
         out=accuracy,
     )

valor_lite/classification/manager.py

@@ -367,6 +367,14 @@ class Evaluator:
             )
         ]
 
+        metrics[MetricType.Accuracy] = [
+            Accuracy(
+                value=accuracy.astype(float).tolist(),
+                score_thresholds=score_thresholds,
+                hardmax=hardmax,
+            )
+        ]
+
         for label_idx, label in self.index_to_label.items():
 
             kwargs = {
@@ -401,12 +409,6 @@ class Evaluator:
                     **kwargs,
                 )
             )
-            metrics[MetricType.Accuracy].append(
-                Accuracy(
-                    value=accuracy[:, label_idx].astype(float).tolist(),
-                    **kwargs,
-                )
-            )
             metrics[MetricType.F1].append(
                 F1(
                     value=f1_score[:, label_idx].astype(float).tolist(),

valor_lite/classification/metric.py

@@ -158,24 +158,23 @@ class Recall(_ThresholdValue):
     pass
 
 
-class Accuracy(_ThresholdValue):
+class F1(_ThresholdValue):
     """
-    Accuracy metric for a specific class label.
+    F1 score for a specific class label.
 
-    This class calculates the accuracy at various score thresholds for a binary
-    classification task. Accuracy is defined as the ratio of the sum of true positives and
-    true negatives over all predictions.
+    This class calculates the F1 score at various score thresholds for a binary
+    classification task.
 
     Attributes
     ----------
     value : list[float]
-        Accuracy values computed at each score threshold.
+        F1 scores computed at each score threshold.
     score_thresholds : list[float]
-        Score thresholds at which the accuracy values are computed.
+        Score thresholds at which the F1 scores are computed.
     hardmax : bool
         Indicates whether hardmax thresholding was used.
     label : str
-        The class label for which the accuracy is computed.
+        The class label for which the F1 score is computed.
 
     Methods
     -------
@@ -188,23 +187,21 @@ class Accuracy(_ThresholdValue):
     pass
 
 
-class F1(_ThresholdValue):
+@dataclass
+class Accuracy:
     """
-    F1 score for a specific class label.
+    Multiclass accuracy metric.
 
-    This class calculates the F1 score at various score thresholds for a binary
-    classification task.
+    This class calculates the accuracy at various score thresholds.
 
     Attributes
     ----------
     value : list[float]
-        F1 scores computed at each score threshold.
+        Accuracy values computed at each score threshold.
     score_thresholds : list[float]
-        Score thresholds at which the F1 scores are computed.
+        Score thresholds at which the accuracy values are computed.
     hardmax : bool
         Indicates whether hardmax thresholding was used.
-    label : str
-        The class label for which the F1 score is computed.
 
     Methods
     -------
@@ -214,7 +211,22 @@ class F1(_ThresholdValue):
         Converts the instance to a dictionary representation.
     """
 
-    pass
+    value: list[float]
+    score_thresholds: list[float]
+    hardmax: bool
+
+    def to_metric(self) -> Metric:
+        return Metric(
+            type=type(self).__name__,
+            value=self.value,
+            parameters={
+                "score_thresholds": self.score_thresholds,
+                "hardmax": self.hardmax,
+            },
+        )
+
+    def to_dict(self) -> dict:
+        return self.to_metric().to_dict()
 
 
 @dataclass

valor_lite/object_detection/computation.py

@@ -282,6 +282,7 @@ def compute_metrics(
     ],
     NDArray[np.float64],
     NDArray[np.float64],
+    NDArray[np.float64],
 ]:
     """
     Computes Object Detection metrics.
@@ -309,13 +310,15 @@ def compute_metrics(
 
     Returns
     -------
-    tuple[NDArray, NDArray, NDArray, float]
+    tuple[NDArray[np.float64], NDArray[np.float64], NDArray[np.float64], float]
         Average Precision results.
-    tuple[NDArray, NDArray, NDArray, float]
+    tuple[NDArray[np.float64], NDArray[np.float64], NDArray[np.float64], float]
         Average Recall results.
-    np.ndarray
-        Precision, Recall, TP, FP, FN, F1 Score, Accuracy.
-    np.ndarray
+    NDArray[np.float64]
+        Accuracy.
+    NDArray[np.float64]
+        Precision, Recall, TP, FP, FN, F1 Score.
+    NDArray[np.float64]
         Interpolated Precision-Recall Curves.
     """
 
@@ -329,9 +332,10 @@ def compute_metrics(
     elif n_scores == 0:
         raise ValueError("At least one score threshold must be passed.")
 
-    average_precision = np.zeros((n_ious, n_labels))
-    average_recall = np.zeros((n_scores, n_labels))
-    counts = np.zeros((n_ious, n_scores, n_labels, 7))
+    average_precision = np.zeros((n_ious, n_labels), dtype=np.float64)
+    average_recall = np.zeros((n_scores, n_labels), dtype=np.float64)
+    accuracy = np.zeros((n_ious, n_scores), dtype=np.float64)
+    counts = np.zeros((n_ious, n_scores, n_labels, 6), dtype=np.float64)
 
     pd_labels = data[:, 5].astype(np.int32)
     scores = data[:, 6]
@@ -417,14 +421,6 @@ def compute_metrics(
                 out=f1_score,
             )
 
-            accuracy = np.zeros_like(tp_count)
-            np.divide(
-                tp_count,
-                (gt_count + pd_count),
-                where=(gt_count + pd_count) > 1e-9,
-                out=accuracy,
-            )
-
             counts[iou_idx][score_idx] = np.concatenate(
                 (
                     tp_count[:, np.newaxis],
@@ -433,11 +429,18 @@ def compute_metrics(
                     precision[:, np.newaxis],
                     recall[:, np.newaxis],
                     f1_score[:, np.newaxis],
-                    accuracy[:, np.newaxis],
                 ),
                 axis=1,
             )
 
+            # caluculate accuracy
+            total_pd_count = label_metadata[:, 1].sum()
+            accuracy[iou_idx, score_idx] = (
+                (tp_count.sum() / total_pd_count)
+                if total_pd_count > 1e-9
+                else 0.0
+            )
+
             # calculate recall for AR
             average_recall[score_idx] += recall
 
@@ -552,6 +555,7 @@ def compute_metrics(
     return (
         ap_results,
         ar_results,
+        accuracy,
         counts,
         pr_curve,
     )

valor_lite/object_detection/manager.py

@@ -506,6 +506,7 @@ class Evaluator:
                 average_recall_averaged_over_scores,
                 mean_average_recall_averaged_over_scores,
             ),
+            accuracy,
             precision_recall,
             pr_curves,
         ) = compute_metrics(
@@ -593,6 +594,16 @@ class Evaluator:
             )
         ]
 
+        metrics[MetricType.Accuracy] = [
+            Accuracy(
+                value=float(accuracy[iou_idx, score_idx]),
+                iou_threshold=iou_thresholds[iou_idx],
+                score_threshold=score_thresholds[score_idx],
+            )
+            for iou_idx in range(accuracy.shape[0])
+            for score_idx in range(accuracy.shape[1])
+        ]
+
         metrics[MetricType.PrecisionRecallCurve] = [
             PrecisionRecallCurve(
                 precisions=pr_curves[iou_idx, label_idx, :, 0]
@@ -650,12 +661,6 @@ class Evaluator:
                             **kwargs,
                         )
                     )
-                    metrics[MetricType.Accuracy].append(
-                        Accuracy(
-                            value=float(row[6]),
-                            **kwargs,
-                        )
-                    )
 
         if as_dict:
             return {

valor_lite/object_detection/metric.py

@@ -160,9 +160,9 @@ class Recall(_ClassMetric):
     pass
 
 
-class Accuracy(_ClassMetric):
+class F1(_ClassMetric):
     """
-    Accuracy metric for a specific class label in object detection.
+    F1 score for a specific class label in object detection.
 
     This class encapsulates a metric value for a particular class label,
     along with the associated Intersection over Union (IoU) threshold and
@@ -190,20 +190,18 @@ class Accuracy(_ClassMetric):
     pass
 
 
-class F1(_ClassMetric):
+@dataclass
+class Accuracy:
     """
-    F1 score for a specific class label in object detection.
+    Accuracy metric for the object detection task type.
 
-    This class encapsulates a metric value for a particular class label,
-    along with the associated Intersection over Union (IoU) threshold and
-    confidence score threshold.
+    This class encapsulates a metric value at a specific Intersection
+    over Union (IoU) threshold and confidence score threshold.
 
     Attributes
     ----------
     value : float
         The metric value.
-    label : str
-        The class label for which the metric is calculated.
     iou_threshold : float
         The IoU threshold used to determine matches between predicted and ground truth boxes.
     score_threshold : float
@@ -217,7 +215,22 @@ class F1(_ClassMetric):
         Converts the instance to a dictionary representation.
     """
 
-    pass
+    value: float
+    iou_threshold: float
+    score_threshold: float
+
+    def to_metric(self) -> Metric:
+        return Metric(
+            type=type(self).__name__,
+            value=self.value,
+            parameters={
+                "iou_threshold": self.iou_threshold,
+                "score_threshold": self.score_threshold,
+            },
+        )
+
+    def to_dict(self) -> dict:
+        return self.to_metric().to_dict()
 
 
 @dataclass

{valor_lite-0.33.11.dist-info → valor_lite-0.33.13.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: valor-lite
-Version: 0.33.11
+Version: 0.33.13
 Summary: Compute valor metrics locally.
 License: MIT License
         

{valor_lite-0.33.11.dist-info → valor_lite-0.33.13.dist-info}/RECORD

@@ -3,22 +3,22 @@ valor_lite/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 valor_lite/schemas.py,sha256=r4cC10w1xYsA785KmGE4ePeOX3wzEs846vT7QAiVg_I,293
 valor_lite/classification/__init__.py,sha256=2wmmziIzUATm7MbmAcPNLXrEX5l4oeD7XBwPd9bWM3Q,506
 valor_lite/classification/annotation.py,sha256=0aUOvcwBAZgiNOJuyh-pXyNTG7vP7r8CUfnU3OmpUwQ,1113
-valor_lite/classification/computation.py,sha256=qd9K7CcSGmMm_7shfX47_ZIuB-uE2LLiLMZSS_3NJTk,12093
-valor_lite/classification/manager.py,sha256=7NKk4syQHH5hBEUDWTD0zIFkJSNdOMzJn8a8GzfBnDc,23205
-valor_lite/classification/metric.py,sha256=m9_zD82YGl0QhuMql943YNKg67NZ6bsrR8ggs6_JZms,11728
+valor_lite/classification/computation.py,sha256=pMePRFKCikYiGDgR-ZB8TmrzAts5ZIz4EywCT-XL42g,12100
+valor_lite/classification/manager.py,sha256=fwb5z84SzgJ-ud1kTY3oYbUJLbA7R0cdWqqcaAIUcWs,23222
+valor_lite/classification/metric.py,sha256=JjY9x6Sq1Hr_2agGnyT9EhVI5wXKQcMmEwxIK32yhGw,11903
 valor_lite/object_detection/__init__.py,sha256=PiKfemo8FkZRzBhPSjhil8ahGURLy0Vk_iV25CB4UBU,1139
 valor_lite/object_detection/annotation.py,sha256=o6VfiRobiB0ljqsNBLAYMXgi32RSIR7uTA-dgxq6zBI,8248
-valor_lite/object_detection/computation.py,sha256=ZW83XT-aemRg-5ZdISmrj0bRD9wWmYCU3gkSlfXlNZc,27747
-valor_lite/object_detection/manager.py,sha256=vb4JpynNF0JcnFwNmReFjls9UGAquigN2hpEbG89J04,38991
-valor_lite/object_detection/metric.py,sha256=tHRVnpBqw_w1VwnNkTCmu1yv7Max9FRlf5uh0wYew4s,24046
+valor_lite/object_detection/computation.py,sha256=dbwyBgRQqG47R8FINd0vQq10b85rsq0jZH4M81KQT24,28017
+valor_lite/object_detection/manager.py,sha256=gjKpytNldF51V_xUktJHrLRDQme-AkZ3HpiL8uMnYJY,39156
+valor_lite/object_detection/metric.py,sha256=SS3U-HV3QgHoN3hcY2DmLl5GzK4KyvC78vjXTIa7XAU,24330
 valor_lite/semantic_segmentation/__init__.py,sha256=IdarTHKUuUMDvMBmInQu12Mm_NMCbql6Hf0nL5b56Ak,424
 valor_lite/semantic_segmentation/annotation.py,sha256=CujYFdHS3fgr4Y7mEDs_u1XBmbPJzNU2CdqvjCT_d_A,2938
 valor_lite/semantic_segmentation/computation.py,sha256=iJkEmTNmw9HwQCxSnpJkQsAdVcFriGhhu_WMks6D7tU,5122
 valor_lite/semantic_segmentation/manager.py,sha256=aJk6edWZWKqrzl6hVmEUSZVYhHLuyihxWgAIXsCXkZ0,17361
 valor_lite/semantic_segmentation/metric.py,sha256=Y8M3z92SaABEe9TwBUN37TFsh9DR5WoIxO-TfXVwz8I,6289
 valor_lite/text_generation/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-valor_lite-0.33.11.dist-info/LICENSE,sha256=M0L53VuwfEEqezhHb7NPeYcO_glw7-k4DMLZQ3eRN64,1068
-valor_lite-0.33.11.dist-info/METADATA,sha256=QniFV86iMnaqPtJElufV4tkF3-kI1sS6EXKRzupWavc,5632
-valor_lite-0.33.11.dist-info/WHEEL,sha256=GV9aMThwP_4oNCtvEC2ec3qUYutgWeAzklro_0m4WJQ,91
-valor_lite-0.33.11.dist-info/top_level.txt,sha256=9ujykxSwpl2Hu0_R95UQTR_l07k9UUTSdrpiqmq6zc4,11
-valor_lite-0.33.11.dist-info/RECORD,,
+valor_lite-0.33.13.dist-info/LICENSE,sha256=M0L53VuwfEEqezhHb7NPeYcO_glw7-k4DMLZQ3eRN64,1068
+valor_lite-0.33.13.dist-info/METADATA,sha256=4WsoZ-i3KETNN5d1F333ZJvhGPj1tiGEAzSTjrOW7yk,5632
+valor_lite-0.33.13.dist-info/WHEEL,sha256=OVMc5UfuAQiSplgO0_WdW7vXVGAt9Hdd6qtN4HotdyA,91
+valor_lite-0.33.13.dist-info/top_level.txt,sha256=9ujykxSwpl2Hu0_R95UQTR_l07k9UUTSdrpiqmq6zc4,11
+valor_lite-0.33.13.dist-info/RECORD,,

{valor_lite-0.33.11.dist-info → valor_lite-0.33.13.dist-info}/WHEEL

@@ -1,5 +1,5 @@
 Wheel-Version: 1.0
-Generator: setuptools (75.1.0)
+Generator: setuptools (75.2.0)
 Root-Is-Purelib: true
 Tag: py3-none-any