easy_rec/python/core/easyrec_metrics/distribute_metrics_impl_pai.py [798:843]:
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
              name='pr_auc_increment'),
          name='interpolate_pr_auc')

    def compute_auc(tp, fn, tn, fp, name):
      """Computes the roc-auc or pr-auc based on confusion counts."""
      if curve == 'PR':
        if summation_method == 'trapezoidal':
          logging.warning(
              'Trapezoidal rule is known to produce incorrect PR-AUCs; '
              'please switch to "careful_interpolation" instead.')
        elif summation_method == 'careful_interpolation':
          # This one is a bit tricky and is handled separately.
          return interpolate_pr_auc(tp, fp, fn)
      rec = math_ops.div(tp + epsilon, tp + fn + epsilon)
      if curve == 'ROC':
        fp_rate = math_ops.div(fp, fp + tn + epsilon)
        x = fp_rate
        y = rec
      else:  # curve == 'PR'.
        prec = math_ops.div(tp + epsilon, tp + fp + epsilon)
        x = rec
        y = prec
      if summation_method in ('trapezoidal', 'careful_interpolation'):
        # Note that the case ('PR', 'careful_interpolation') has been handled
        # above.
        return math_ops.reduce_sum(
            math_ops.multiply(x[:num_thresholds - 1] - x[1:],
                              (y[:num_thresholds - 1] + y[1:]) / 2.),
            name=name)
      elif summation_method == 'minoring':
        return math_ops.reduce_sum(
            math_ops.multiply(x[:num_thresholds - 1] - x[1:],
                              math_ops.minimum(y[:num_thresholds - 1], y[1:])),
            name=name)
      elif summation_method == 'majoring':
        return math_ops.reduce_sum(
            math_ops.multiply(x[:num_thresholds - 1] - x[1:],
                              math_ops.maximum(y[:num_thresholds - 1], y[1:])),
            name=name)
      else:
        raise ValueError('Invalid summation_method: %s' % summation_method)

    # sum up the areas of all the trapeziums
    def compute_auc_value(_, values):
      return compute_auc(values['tp'], values['fn'], values['tn'], values['fp'],
                         'value')
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -



easy_rec/python/core/easyrec_metrics/distribute_metrics_impl_tf.py [804:849]:
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
              name='pr_auc_increment'),
          name='interpolate_pr_auc')

    def compute_auc(tp, fn, tn, fp, name):
      """Computes the roc-auc or pr-auc based on confusion counts."""
      if curve == 'PR':
        if summation_method == 'trapezoidal':
          logging.warning(
              'Trapezoidal rule is known to produce incorrect PR-AUCs; '
              'please switch to "careful_interpolation" instead.')
        elif summation_method == 'careful_interpolation':
          # This one is a bit tricky and is handled separately.
          return interpolate_pr_auc(tp, fp, fn)
      rec = math_ops.div(tp + epsilon, tp + fn + epsilon)
      if curve == 'ROC':
        fp_rate = math_ops.div(fp, fp + tn + epsilon)
        x = fp_rate
        y = rec
      else:  # curve == 'PR'.
        prec = math_ops.div(tp + epsilon, tp + fp + epsilon)
        x = rec
        y = prec
      if summation_method in ('trapezoidal', 'careful_interpolation'):
        # Note that the case ('PR', 'careful_interpolation') has been handled
        # above.
        return math_ops.reduce_sum(
            math_ops.multiply(x[:num_thresholds - 1] - x[1:],
                              (y[:num_thresholds - 1] + y[1:]) / 2.),
            name=name)
      elif summation_method == 'minoring':
        return math_ops.reduce_sum(
            math_ops.multiply(x[:num_thresholds - 1] - x[1:],
                              math_ops.minimum(y[:num_thresholds - 1], y[1:])),
            name=name)
      elif summation_method == 'majoring':
        return math_ops.reduce_sum(
            math_ops.multiply(x[:num_thresholds - 1] - x[1:],
                              math_ops.maximum(y[:num_thresholds - 1], y[1:])),
            name=name)
      else:
        raise ValueError('Invalid summation_method: %s' % summation_method)

    # sum up the areas of all the trapeziums
    def compute_auc_value(_, values):
      return compute_auc(values['tp'], values['fn'], values['tn'], values['fp'],
                         'value')
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -