Module: tfr.keras.metrics

Keras metrics in TF-Ranking.

Note: For metrics that compute a ranking, ties are broken randomly. This means that metrics may be stochastic if items with equal scores are provided.

Warning: Some metrics (e.g. Recall or MRR) are not well-defined when there are no relevant items (e.g. if y_true has a row of only zeroes). For these cases, the TF-Ranking metrics will evaluate to 0.

Classes

class ARPMetric: Average relevance position (ARP).

class AlphaDCGMetric: Alpha discounted cumulative gain (alphaDCG).

class DCGMetric: Discounted cumulative gain (DCG).

class HitsMetric: Hits@k metric.

class MRRMetric: Mean reciprocal rank (MRR).

class MeanAveragePrecisionMetric: Mean average precision (MAP).

class NDCGMetric: Normalized discounted cumulative gain (NDCG).

class OPAMetric: Ordered pair accuracy (OPA).

class PrecisionIAMetric: Precision-IA@k (Pre-IA@k).

class PrecisionMetric: Precision@k (P@k).

class RankingMetricKey: Ranking metric key strings.

class RecallMetric: Recall@k (R@k).

Functions

default_keras_metrics(...): Returns a list of ranking metrics.

get(...): Factory method to get a list of ranking metrics.

Other Members
absolute_import	Instance of __future__._Feature
division	Instance of __future__._Feature
print_function	Instance of __future__._Feature