Add Regression AUC (RAUC) metrics

Summary:
Implement regression AUC metrics. Regression AUC is an extension of classification AUC. See Section 4.1.1 in https://arxiv.org/ftp/arxiv/papers/1205/1205.2618.pdf for related discussions.

On a high level, regression AUC is an extension of the traditional AUC for classification through the probabilistic interpretation: the area under the curve is equal to the probability that a classifier will rank a randomly chosen positive instance higher than a randomly chosen negative one.

We utilize merge sort to optimize time complexity to O(nlog(n)).

Reviewed By: zainhuda

Differential Revision: D53377225

fbshipit-source-id: 7dfccbddf9f17a6881c6fd00f9614466c603514d

Author: Z Zhou

torchrec/metrics/metric_module.py

@@ -39,6 +39,7 @@
 from torchrec.metrics.multiclass_recall import MulticlassRecallMetric
 from torchrec.metrics.ndcg import NDCGMetric
 from torchrec.metrics.ne import NEMetric
+from torchrec.metrics.rauc import RAUCMetric
 from torchrec.metrics.rec_metric import RecMetric, RecMetricList
 from torchrec.metrics.recall_session import RecallSessionMetric
 from torchrec.metrics.scalar import ScalarMetric
@@ -58,6 +59,7 @@
     RecMetricEnum.CALIBRATION: CalibrationMetric,
     RecMetricEnum.AUC: AUCMetric,
     RecMetricEnum.AUPRC: AUPRCMetric,
+    RecMetricEnum.RAUC: RAUCMetric,
     RecMetricEnum.MSE: MSEMetric,
     RecMetricEnum.MAE: MAEMetric,
     RecMetricEnum.MULTICLASS_RECALL: MulticlassRecallMetric,

torchrec/metrics/metrics_config.py

@@ -23,6 +23,7 @@ class RecMetricEnum(RecMetricEnumBase):
     CTR = "ctr"
     AUC = "auc"
     AUPRC = "auprc"
+    RAUC = "rauc"
     CALIBRATION = "calibration"
     MSE = "mse"
     MAE = "mae"

torchrec/metrics/metrics_namespace.py

@@ -49,8 +49,10 @@ class MetricName(MetricNameBase):
     RMSE = "rmse"
     AUC = "auc"
     AUPRC = "auprc"
+    RAUC = "rauc"
     GROUPED_AUC = "grouped_auc"
     GROUPED_AUPRC = "grouped_auprc"
+    GROUPED_RAUC = "grouped_rauc"
     RECALL_SESSION_LEVEL = "recall_session_level"
     MULTICLASS_RECALL = "multiclass_recall"
     WEIGHTED_AVG = "weighted_avg"
@@ -76,6 +78,7 @@ class MetricNamespace(MetricNamespaceBase):
     MSE = "mse"
     AUC = "auc"
     AUPRC = "auprc"
+    RAUC = "rauc"
     MAE = "mae"
     ACCURACY = "accuracy"