std::vector UpdateNumExamplesComparatorDirect()

in tensorflow_data_validation/anomalies/dataset_constraints_util.cc [46:126]
73 lines of code
13 McCabe index (conditional complexity)

std::vector<Description> UpdateNumExamplesComparatorDirect(
    const DatasetStatsView& stats, const DatasetComparatorType comparator_type,
    tensorflow::metadata::v0::NumericValueComparator* comparator) {
  if (!comparator->has_min_fraction_threshold() &&
      !comparator->has_max_fraction_threshold()) {
    return {};
  }
  double num_examples = stats.GetNumExamples();
  // ValidateFeatureStatistics does not attempt to detect anomalies in
  // datasets that have num_examples == 0. Check that here.
  CHECK(num_examples > 0.0)
      << "Invalid input. Num examples must be greater than "
         "0.";

  const absl::optional<DatasetStatsView> control_stats =
      ((comparator_type == DatasetComparatorType::DRIFT)
           ? stats.GetPreviousSpan()
           : stats.GetPreviousVersion());
  if (!control_stats) {
    return {};
  }
  const string control_name =
      (comparator_type == DatasetComparatorType::DRIFT ? "previous span"
                                                       : "previous version");

  std::vector<Description> descriptions = {};
  double control_num_examples = control_stats->GetNumExamples();
  CHECK(control_num_examples >= 0.0) << "Invalid input. Control num examples "
                                      "must not be negative";

  double num_examples_ratio;
  if (control_num_examples != 0.0) {
    num_examples_ratio = num_examples / control_num_examples;
  }

  // TODO(b/138589350): Check for possible case of ratio == 1.0 but num_examples
  // != control_num_examples.
  if (comparator->has_max_fraction_threshold()) {
    double max_threshold = comparator->max_fraction_threshold();
    if (control_num_examples == 0.0) {
      comparator->clear_max_fraction_threshold();
      descriptions.push_back(
          {tensorflow::metadata::v0::AnomalyInfo::COMPARATOR_HIGH_NUM_EXAMPLES,
           absl::StrCat("High num examples in current dataset versus the ",
                        control_name, ", which has 0."),
           absl::StrCat("The ", control_name,
                        " has 0 examples, so there is a high number of "
                        "examples in the current dataset versus the ",
                        control_name, ".")});
    } else if (num_examples_ratio > max_threshold) {
      comparator->set_max_fraction_threshold(num_examples_ratio);
      descriptions.push_back(
          {tensorflow::metadata::v0::AnomalyInfo::COMPARATOR_HIGH_NUM_EXAMPLES,
           absl::StrCat("High num examples in current dataset versus the ",
                        control_name, "."),
           absl::StrCat(
               "The ratio of num examples in the current dataset versus the ",
               control_name, " is ", absl::SixDigits(num_examples_ratio),
               " (up to six significant digits), which is above the "
               "threshold ",
               absl::SixDigits(max_threshold), ".")});
    }
  }
  if (comparator->has_min_fraction_threshold()) {
    double min_threshold = comparator->min_fraction_threshold();
    if (control_num_examples != 0.0 &&
        num_examples_ratio < comparator->min_fraction_threshold()) {
      comparator->set_min_fraction_threshold(num_examples_ratio);
      descriptions.push_back(
          {tensorflow::metadata::v0::AnomalyInfo::COMPARATOR_LOW_NUM_EXAMPLES,
           absl::StrCat("Low num examples in current dataset versus the ",
                        control_name, "."),
           absl::StrCat(
               "The ratio of num examples in the current dataset versus the ",
               control_name, " is ", absl::SixDigits(num_examples_ratio),
               " (up to six significant digits), which is below the threshold ",
               absl::SixDigits(min_threshold), ".")});
    }
  }
  return descriptions;
}