Compute numerical statistics

Stay organized with collections Save and categorize content based on your preferences.

You can determine minimum, maximum, and quantile values for an individual BigQuery column. To calculate these values, you configure a DlpJob, setting the NumericalStatsConfig privacy metric to the name of the column to scan. When you run the job, Cloud DLP computes statistics for the given column, returning its results in the NumericalStatsResult object.

Explore further

For detailed documentation that includes this code sample, see the following:

Code sample


To learn how to install and use the client library for Cloud DLP, see Cloud DLP client libraries.

using Google.Api.Gax.ResourceNames;
using Google.Cloud.Dlp.V2;
using Google.Cloud.PubSub.V1;
using Newtonsoft.Json;
using System;
using System.Collections.Generic;
using System.Linq;
using System.Threading;
using System.Threading.Tasks;
using static Google.Cloud.Dlp.V2.Action.Types;
using static Google.Cloud.Dlp.V2.PrivacyMetric.Types;

public class RiskAnalysisCreateNumericalStats
    public static AnalyzeDataSourceRiskDetails.Types.NumericalStatsResult NumericalStats(
        string callingProjectId,
        string tableProjectId,
        string datasetId,
        string tableId,
        string topicId,
        string subscriptionId,
        string columnName)
        var dlp = DlpServiceClient.Create();

        // Construct + submit the job
        var config = new RiskAnalysisJobConfig
            PrivacyMetric = new PrivacyMetric
                NumericalStatsConfig = new NumericalStatsConfig
                    Field = new FieldId { Name = columnName }
            SourceTable = new BigQueryTable
                ProjectId = tableProjectId,
                DatasetId = datasetId,
                TableId = tableId
            Actions =
                    new Google.Cloud.Dlp.V2.Action
                        PubSub = new PublishToPubSub
                            Topic = $"projects/{callingProjectId}/topics/{topicId}"

        var submittedJob = dlp.CreateDlpJob(
            new CreateDlpJobRequest
                ParentAsProjectName = new ProjectName(callingProjectId),
                RiskJob = config

        // Listen to pub/sub for the job
        var subscriptionName = new SubscriptionName(callingProjectId, subscriptionId);
        var subscriber = SubscriberClient.CreateAsync(

        // SimpleSubscriber runs your message handle function on multiple
        // threads to maximize throughput.
        var done = new ManualResetEventSlim(false);
        subscriber.StartAsync((PubsubMessage message, CancellationToken cancel) =>
            if (message.Attributes["DlpJobName"] == submittedJob.Name)
                Thread.Sleep(500); // Wait for DLP API results to become consistent
                return Task.FromResult(SubscriberClient.Reply.Ack);
                return Task.FromResult(SubscriberClient.Reply.Nack);

        done.Wait(TimeSpan.FromMinutes(10)); // 10 minute timeout; may not work for large jobs

        // Process results
        var resultJob = dlp.GetDlpJob(
            new GetDlpJobRequest
                DlpJobName = DlpJobName.Parse(submittedJob.Name)

        var result = resultJob.RiskDetails.NumericalStatsResult;

        // 'UnpackValue(x)' is a prettier version of 'x.toString()'
        Console.WriteLine($"Value Range: [{UnpackValue(result.MinValue)}, {UnpackValue(result.MaxValue)}]");
        var lastValue = string.Empty;
        for (var quantile = 0; quantile < result.QuantileValues.Count; quantile++)
            var currentValue = UnpackValue(result.QuantileValues[quantile]);
            if (lastValue != currentValue)
                Console.WriteLine($"Value at {quantile + 1}% quantile: {currentValue}");
            lastValue = currentValue;

        return result;

    public static string UnpackValue(Value protoValue)
        var jsonValue = JsonConvert.DeserializeObject<Dictionary<string, object>>(protoValue.ToString());
        return jsonValue.Values.ElementAt(0).ToString();


To learn how to install and use the client library for Cloud DLP, see Cloud DLP client libraries.

import (

	dlp ""
	dlppb ""

// riskNumerical computes the numerical risk of the given column.
func riskNumerical(w io.Writer, projectID, dataProject, pubSubTopic, pubSubSub, datasetID, tableID, columnName string) error {
	// projectID := "my-project-id"
	// dataProject := "bigquery-public-data"
	// pubSubTopic := "dlp-risk-sample-topic"
	// pubSubSub := "dlp-risk-sample-sub"
	// datasetID := "nhtsa_traffic_fatalities"
	// tableID := "accident_2015"
	// columnName := "state_number"
	ctx := context.Background()
	client, err := dlp.NewClient(ctx)
	if err != nil {
		return fmt.Errorf("dlp.NewClient: %v", err)
	// Create a PubSub Client used to listen for when the inspect job finishes.
	pubsubClient, err := pubsub.NewClient(ctx, projectID)
	if err != nil {
		return fmt.Errorf("Error creating PubSub client: %v", err)
	defer pubsubClient.Close()

	// Create a PubSub subscription we can use to listen for messages.
	s, err := setupPubSub(projectID, pubSubTopic, pubSubSub)
	if err != nil {
		return fmt.Errorf("setupPubSub: %v", err)

	// topic is the PubSub topic string where messages should be sent.
	topic := "projects/" + projectID + "/topics/" + pubSubTopic

	// Create a configured request.
	req := &dlppb.CreateDlpJobRequest{
		Parent: fmt.Sprintf("projects/%s/locations/global", projectID),
		Job: &dlppb.CreateDlpJobRequest_RiskJob{
			RiskJob: &dlppb.RiskAnalysisJobConfig{
				// PrivacyMetric configures what to compute.
				PrivacyMetric: &dlppb.PrivacyMetric{
					Type: &dlppb.PrivacyMetric_NumericalStatsConfig_{
						NumericalStatsConfig: &dlppb.PrivacyMetric_NumericalStatsConfig{
							Field: &dlppb.FieldId{
								Name: columnName,
				// SourceTable describes where to find the data.
				SourceTable: &dlppb.BigQueryTable{
					ProjectId: dataProject,
					DatasetId: datasetID,
					TableId:   tableID,
				// Send a message to PubSub using Actions.
				Actions: []*dlppb.Action{
						Action: &dlppb.Action_PubSub{
							PubSub: &dlppb.Action_PublishToPubSub{
								Topic: topic,
	// Create the risk job.
	j, err := client.CreateDlpJob(ctx, req)
	if err != nil {
		return fmt.Errorf("CreateDlpJob: %v", err)
	fmt.Fprintf(w, "Created job: %v\n", j.GetName())

	// Wait for the risk job to finish by waiting for a PubSub message.
	// This only waits for 10 minutes. For long jobs, consider using a truly
	// asynchronous execution model such as Cloud Functions.
	ctx, cancel := context.WithTimeout(ctx, 10*time.Minute)
	defer cancel()
	err = s.Receive(ctx, func(ctx context.Context, msg *pubsub.Message) {
		// If this is the wrong job, do not process the result.
		if msg.Attributes["DlpJobName"] != j.GetName() {
		time.Sleep(500 * time.Millisecond)
		resp, err := client.GetDlpJob(ctx, &dlppb.GetDlpJobRequest{
			Name: j.GetName(),
		if err != nil {
			fmt.Fprintf(w, "GetDlpJob: %v", err)
		n := resp.GetRiskDetails().GetNumericalStatsResult()
		fmt.Fprintf(w, "Value range: [%v, %v]\n", n.GetMinValue(), n.GetMaxValue())
		var tmp string
		for p, v := range n.GetQuantileValues() {
			if v.String() != tmp {
				fmt.Fprintf(w, "Value at %v quantile: %v\n", p, v)
				tmp = v.String()
		// Stop listening for more messages.
	if err != nil {
		return fmt.Errorf("Recieve: %v", err)
	return nil


To learn how to install and use the client library for Cloud DLP, see Cloud DLP client libraries.

import java.util.concurrent.ExecutionException;
import java.util.concurrent.TimeUnit;
import java.util.concurrent.TimeoutException;

class RiskAnalysisNumericalStats {

  public static void main(String[] args) throws Exception {
    // TODO(developer): Replace these variables before running the sample.
    String projectId = "your-project-id";
    String datasetId = "your-bigquery-dataset-id";
    String tableId = "your-bigquery-table-id";
    String topicId = "pub-sub-topic";
    String subscriptionId = "pub-sub-subscription";
    numericalStatsAnalysis(projectId, datasetId, tableId, topicId, subscriptionId);

  public static void numericalStatsAnalysis(
      String projectId, String datasetId, String tableId, String topicId, String subscriptionId)
      throws ExecutionException, InterruptedException, IOException {

    // Initialize client that will be used to send requests. This client only needs to be created
    // once, and can be reused for multiple requests. After completing all of your requests, call
    // the "close" method on the client to safely clean up any remaining background resources.
    try (DlpServiceClient dlpServiceClient = DlpServiceClient.create()) {

      // Specify the BigQuery table to analyze
      BigQueryTable bigQueryTable =

      // This represents the name of the column to analyze, which must contain numerical data
      String columnName = "Age";

      // Configure the privacy metric for the job
      FieldId fieldId = FieldId.newBuilder().setName(columnName).build();
      NumericalStatsConfig numericalStatsConfig =
      PrivacyMetric privacyMetric =

      // Create action to publish job status notifications over Google Cloud Pub/Sub
      ProjectTopicName topicName = ProjectTopicName.of(projectId, topicId);
      PublishToPubSub publishToPubSub =
      Action action = Action.newBuilder().setPubSub(publishToPubSub).build();

      // Configure the risk analysis job to perform
      RiskAnalysisJobConfig riskAnalysisJobConfig =

      CreateDlpJobRequest createDlpJobRequest =
              .setParent(LocationName.of(projectId, "global").toString())

      // Send the request to the API using the client
      DlpJob dlpJob = dlpServiceClient.createDlpJob(createDlpJobRequest);

      // Set up a Pub/Sub subscriber to listen on the job completion status
      final SettableApiFuture<Boolean> done = SettableApiFuture.create();

      ProjectSubscriptionName subscriptionName =
          ProjectSubscriptionName.of(projectId, subscriptionId);

      MessageReceiver messageHandler =
          (PubsubMessage pubsubMessage, AckReplyConsumer ackReplyConsumer) -> {
            handleMessage(dlpJob, done, pubsubMessage, ackReplyConsumer);
      Subscriber subscriber = Subscriber.newBuilder(subscriptionName, messageHandler).build();

      // Wait for job completion semi-synchronously
      // For long jobs, consider using a truly asynchronous execution model such as Cloud Functions
      try {
        done.get(15, TimeUnit.MINUTES);
      } catch (TimeoutException e) {
        System.out.println("Job was not completed after 15 minutes.");
      } finally {

      // Build a request to get the completed job
      GetDlpJobRequest getDlpJobRequest =

      // Retrieve completed job status
      DlpJob completedJob = dlpServiceClient.getDlpJob(getDlpJobRequest);
      System.out.println("Job status: " + completedJob.getState());

      // Get the result and parse through and process the information
      NumericalStatsResult result = completedJob.getRiskDetails().getNumericalStatsResult();

          "Value range : [%.3f, %.3f]\n",
          result.getMinValue().getFloatValue(), result.getMaxValue().getFloatValue());

      int percent = 1;
      Double lastValue = null;
      for (Value quantileValue : result.getQuantileValuesList()) {
        Double currentValue = quantileValue.getFloatValue();
        if (lastValue == null || !lastValue.equals(currentValue)) {
          System.out.printf("Value at %s %% quantile : %.3f", percent, currentValue);
        lastValue = currentValue;

  // handleMessage injects the job and settableFuture into the message reciever interface
  private static void handleMessage(
      DlpJob job,
      SettableApiFuture<Boolean> done,
      PubsubMessage pubsubMessage,
      AckReplyConsumer ackReplyConsumer) {
    String messageAttribute = pubsubMessage.getAttributesMap().get("DlpJobName");
    if (job.getName().equals(messageAttribute)) {
    } else {


To learn how to install and use the client library for Cloud DLP, see Cloud DLP client libraries.

// Import the Google Cloud client libraries
const DLP = require('@google-cloud/dlp');
const {PubSub} = require('@google-cloud/pubsub');

// Instantiates clients
const dlp = new DLP.DlpServiceClient();
const pubsub = new PubSub();

// The project ID to run the API call under
// const projectId = 'my-project';

// The project ID the table is stored under
// This may or (for public datasets) may not equal the calling project ID
// const tableProjectId = 'my-project';

// The ID of the dataset to inspect, e.g. 'my_dataset'
// const datasetId = 'my_dataset';

// The ID of the table to inspect, e.g. 'my_table'
// const tableId = 'my_table';

// The name of the column to compute risk metrics for, e.g. 'age'
// Note that this column must be a numeric data type
// const columnName = 'firstName';

// The name of the Pub/Sub topic to notify once the job completes
// TODO(developer): create a Pub/Sub topic to use for this
// const topicId = 'MY-PUBSUB-TOPIC'

// The name of the Pub/Sub subscription to use when listening for job
// completion notifications
// TODO(developer): create a Pub/Sub subscription to use for this
// const subscriptionId = 'MY-PUBSUB-SUBSCRIPTION'

async function numericalRiskAnalysis() {
  const sourceTable = {
    projectId: tableProjectId,
    datasetId: datasetId,
    tableId: tableId,

  // Construct request for creating a risk analysis job
  const request = {
    parent: `projects/${projectId}/locations/global`,
    riskJob: {
      privacyMetric: {
        numericalStatsConfig: {
          field: {
            name: columnName,
      sourceTable: sourceTable,
      actions: [
          pubSub: {
            topic: `projects/${projectId}/topics/${topicId}`,

  // Create helper function for unpacking values
  const getValue = obj => obj[Object.keys(obj)[0]];

  // Run risk analysis job
  const [topicResponse] = await pubsub.topic(topicId).get();
  const subscription = await topicResponse.subscription(subscriptionId);
  const [jobsResponse] = await dlp.createDlpJob(request);
  const jobName =;
  // Watch the Pub/Sub topic until the DLP job finishes
  await new Promise((resolve, reject) => {
    const messageHandler = message => {
      if (message.attributes && message.attributes.DlpJobName === jobName) {
        subscription.removeListener('message', messageHandler);
        subscription.removeListener('error', errorHandler);
      } else {

    const errorHandler = err => {
      subscription.removeListener('message', messageHandler);
      subscription.removeListener('error', errorHandler);

    subscription.on('message', messageHandler);
    subscription.on('error', errorHandler);
  setTimeout(() => {
    console.log(' Waiting for DLP job to fully complete');
  }, 500);
  const [job] = await dlp.getDlpJob({name: jobName});
  const results = job.riskDetails.numericalStatsResult;

    `Value Range: [${getValue(results.minValue)}, ${getValue(

  // Print unique quantile values
  let tempValue = null;
  results.quantileValues.forEach((result, percent) => {
    const value = getValue(result);

    // Only print new values
    if (
      tempValue !== value &&
      !(tempValue && tempValue.equals && tempValue.equals(value))
    ) {
      console.log(`Value at ${percent}% quantile: ${value}`);
      tempValue = value;



To learn how to install and use the client library for Cloud DLP, see Cloud DLP client libraries.

use Google\Cloud\Dlp\V2\DlpServiceClient;
use Google\Cloud\Dlp\V2\RiskAnalysisJobConfig;
use Google\Cloud\Dlp\V2\BigQueryTable;
use Google\Cloud\Dlp\V2\DlpJob\JobState;
use Google\Cloud\PubSub\PubSubClient;
use Google\Cloud\Dlp\V2\Action;
use Google\Cloud\Dlp\V2\Action\PublishToPubSub;
use Google\Cloud\Dlp\V2\PrivacyMetric\NumericalStatsConfig;
use Google\Cloud\Dlp\V2\PrivacyMetric;
use Google\Cloud\Dlp\V2\FieldId;

 * Computes risk metrics of a column of numbers in a Google BigQuery table.
 * @param string $callingProjectId  The project ID to run the API call under
 * @param string $dataProjectId     The project ID containing the target Datastore
 * @param string $topicId           The name of the Pub/Sub topic to notify once the job completes
 * @param string $subscriptionId    The name of the Pub/Sub subscription to use when listening for job
 * @param string $datasetId         The ID of the BigQuery dataset to inspect
 * @param string $tableId           The ID of the BigQuery table to inspect
 * @param string $columnName        The name of the column to compute risk metrics for, e.g. "age"
function numerical_stats(
    string $callingProjectId,
    string $dataProjectId,
    string $topicId,
    string $subscriptionId,
    string $datasetId,
    string $tableId,
    string $columnName
): void {
    // Instantiate a client.
    $dlp = new DlpServiceClient([
        'projectId' => $callingProjectId
    $pubsub = new PubSubClient([
        'projectId' => $callingProjectId
    $topic = $pubsub->topic($topicId);

    // Construct risk analysis config
    $columnField = (new FieldId())

    $statsConfig = (new NumericalStatsConfig())

    $privacyMetric = (new PrivacyMetric())

    // Construct items to be analyzed
    $bigqueryTable = (new BigQueryTable())

    // Construct the action to run when job completes
    $pubSubAction = (new PublishToPubSub())

    $action = (new Action())

    // Construct risk analysis job config to run
    $riskJob = (new RiskAnalysisJobConfig())

    // Listen for job notifications via an existing topic/subscription.
    $subscription = $topic->subscription($subscriptionId);

    // Submit request
    $parent = "projects/$callingProjectId/locations/global";
    $job = $dlp->createDlpJob($parent, [
        'riskJob' => $riskJob

    // Poll Pub/Sub using exponential backoff until job finishes
    // Consider using an asynchronous execution model such as Cloud Functions
    $attempt = 1;
    $startTime = time();
    do {
        foreach ($subscription->pull() as $message) {
            if (isset($message->attributes()['DlpJobName']) &&
                $message->attributes()['DlpJobName'] === $job->getName()) {
                // Get the updated job. Loop to avoid race condition with DLP API.
                do {
                    $job = $dlp->getDlpJob($job->getName());
                } while ($job->getState() == JobState::RUNNING);
                break 2; // break from parent do while
        printf('Waiting for job to complete' . PHP_EOL);
        // Exponential backoff with max delay of 60 seconds
        sleep(min(60, pow(2, ++$attempt)));
    } while (time() - $startTime < 600); // 10 minute timeout

    // Helper function to convert Protobuf values to strings
    $valueToString = function ($value) {
        $json = json_decode($value->serializeToJsonString(), true);
        return array_shift($json);

    // Print finding counts
    printf('Job %s status: %s' . PHP_EOL, $job->getName(), JobState::name($job->getState()));
    switch ($job->getState()) {
        case JobState::DONE:
            $results = $job->getRiskDetails()->getNumericalStatsResult();
                'Value range: [%s, %s]' . PHP_EOL,

            // Only print unique values
            $lastValue = null;
            foreach ($results->getQuantileValues() as $percent => $quantileValue) {
                $value = $valueToString($quantileValue);
                if ($value != $lastValue) {
                    printf('Value at %s quantile: %s' . PHP_EOL, $percent, $value);
                    $lastValue = $value;

        case JobState::FAILED:
            printf('Job %s had errors:' . PHP_EOL, $job->getName());
            $errors = $job->getErrors();
            foreach ($errors as $error) {
        case JobState::PENDING:
            printf('Job has not completed. Consider a longer timeout or an asynchronous execution model' . PHP_EOL);
            print('Unexpected job state. Most likely, the job is either running or has not yet started.');


To learn how to install and use the client library for Cloud DLP, see Cloud DLP client libraries.

def numerical_risk_analysis(
    """Uses the Data Loss Prevention API to compute risk metrics of a column
       of numerical data in a Google BigQuery table.
        project: The Google Cloud project id to use as a parent resource.
        table_project_id: The Google Cloud project id where the BigQuery table
            is stored.
        dataset_id: The id of the dataset to inspect.
        table_id: The id of the table to inspect.
        column_name: The name of the column to compute risk metrics for.
        topic_id: The name of the Pub/Sub topic to notify once the job
        subscription_id: The name of the Pub/Sub subscription to use when
            listening for job completion notifications.
        timeout: The number of seconds to wait for a response from the API.

        None; the response from the API is printed to the terminal.
    import concurrent.futures

    # Import the client library.

    # This sample additionally uses Cloud Pub/Sub to receive results from
    # potentially long-running operations.

    # Instantiate a client.
    dlp =

    # Convert the project id into full resource ids.
    topic =, topic_id)
    parent = f"projects/{project}/locations/global"

    # Location info of the BigQuery table.
    source_table = {
        "project_id": table_project_id,
        "dataset_id": dataset_id,
        "table_id": table_id,

    # Tell the API where to send a notification when the job is complete.
    actions = [{"pub_sub": {"topic": topic}}]

    # Configure risk analysis job
    # Give the name of the numeric column to compute risk metrics for
    risk_job = {
        "privacy_metric": {"numerical_stats_config": {"field": {"name": column_name}}},
        "source_table": source_table,
        "actions": actions,

    # Call API to start risk analysis job
    operation = dlp.create_dlp_job(request={"parent": parent, "risk_job": risk_job})

    def callback(message):
        if message.attributes["DlpJobName"] ==
            # This is the message we're looking for, so acknowledge it.

            # Now that the job is done, fetch the results and print them.
            job = dlp.get_dlp_job(request={"name":})
            results = job.risk_details.numerical_stats_result
                "Value Range: [{}, {}]".format(
            prev_value = None
            for percent, result in enumerate(results.quantile_values):
                value = result.integer_value
                if prev_value != value:
                    print("Value at {}% quantile: {}".format(percent, value))
                    prev_value = value
            # This is not the message we're looking for.

    # Create a Pub/Sub client and find the subscription. The subscription is
    # expected to already be listening to the topic.
    subscriber =
    subscription_path = subscriber.subscription_path(project, subscription_id)
    subscription = subscriber.subscribe(subscription_path, callback)

    except concurrent.futures.TimeoutError:
            "No event received before the timeout. Please verify that the "
            "subscription provided is subscribed to the topic provided."

What's next

To search and filter code samples for other Google Cloud products, see the Google Cloud sample browser.