/TensorFlow Python



Defined in tensorflow/contrib/metrics/python/ops/metric_ops.py.

Computes the AUC and asymptotic normally distributed confidence interval.

USAGE NOTE: this approach requires storing all of the predictions and labels for a single evaluation in memory, so it may not be usable when the evaluation batch size and/or the number of evaluation steps is very large.

Computes the area under the ROC curve and its confidence interval using placement values. This has the advantage of being resilient to the distribution of predictions by aggregating across batches, accumulating labels and predictions and performing the final calculation using all of the concatenated values.


  • labels: A Tensor of ground truth labels with the same shape as labels and with values of 0 or 1 whose values are castable to int64.
  • predictions: A Tensor of predictions whose values are castable to float64. Will be flattened into a 1-D Tensor.
  • weights: Optional Tensor whose rank is either 0, or the same rank as labels.
  • alpha: Confidence interval level desired.
  • logit_transformation: A boolean value indicating whether the estimate should be logit transformed prior to calculating the confidence interval. Doing so enforces the restriction that the AUC should never be outside the interval [0,1].
  • metrics_collections: An optional iterable of collections that auc should be added to.
  • updates_collections: An optional iterable of collections that update_op should be added to.
  • name: An optional name for the variable_scope that contains the metric variables.


  • auc: A 1-D Tensor containing the current area-under-curve, lower, and upper confidence interval values.
  • update_op: An operation that concatenates the input labels and predictions to the accumulated values.


  • ValueError: If labels, predictions, and weights have mismatched shapes or if alpha isn't in the range (0,1).

© 2018 The TensorFlow Authors. All rights reserved.
Licensed under the Creative Commons Attribution License 3.0.
Code samples licensed under the Apache 2.0 License.