Method result

tensorflow/python/keras/metrics.py:1819–1852

(self)

        name='interpolate_pr_auc')

  def result(self):
    if (self.curve == metrics_utils.AUCCurve.PR and
        self.summation_method == metrics_utils.AUCSummationMethod.INTERPOLATION
       ):
      # This use case is different and is handled separately.
      return self.interpolate_pr_auc()

    # Set `x` and `y` values for the curves based on `curve` config.
    recall = math_ops.div_no_nan(self.true_positives,
                                 self.true_positives + self.false_negatives)
    if self.curve == metrics_utils.AUCCurve.ROC:
      fp_rate = math_ops.div_no_nan(self.false_positives,
                                    self.false_positives + self.true_negatives)
      x = fp_rate
      y = recall
    else:  # curve == 'PR'.
      precision = math_ops.div_no_nan(
          self.true_positives, self.true_positives + self.false_positives)
      x = recall
      y = precision

    # Find the rectangle heights based on `summation_method`.
    if self.summation_method == metrics_utils.AUCSummationMethod.INTERPOLATION:
      # Note: the case ('PR', 'interpolation') has been handled above.
      heights = (y[:self.num_thresholds - 1] + y[1:]) / 2.
    elif self.summation_method == metrics_utils.AUCSummationMethod.MINORING:
      heights = math_ops.minimum(y[:self.num_thresholds - 1], y[1:])
    else:  # self.summation_method = metrics_utils.AUCSummationMethod.MAJORING:
      heights = math_ops.maximum(y[:self.num_thresholds - 1], y[1:])

    # Sum up the areas of all the rectangles.
    return math_ops.reduce_sum(
        math_ops.multiply(x[:self.num_thresholds - 1] - x[1:], heights),
        name=self.name)

  def reset_states(self):
    K.batch_set_value(