Use Cloud Dataproc, BigQuery, and Apache Spark ML for Machine Learning

The Google BigQuery Connector for Apache Spark allows Data Scientists to blend the power of BigQuery’s seamlessly scalable SQL engine with Apache Spark’s Machine Learning capabilities. In this tutorial, we show how to use Cloud Dataproc, BigQuery, the BigQuery Python client library, and Apache Spark ML to perform machine learning on a dataset.


Use linear regression to build a model of birth weight as a function of five factors:

  1. gestation weeks
  2. mother’s age
  3. father’s age
  4. mother’s weight gain during pregnancy
  5. Apgar score

BigQuery is used to prepare the linear regression input table, which is written to your Google Cloud Platform project. Python running on your local machine is used to query and manage data in BigQuery. The resulting linear regression table is accessed in Apache Spark, and Spark ML is used to build and evaluate the model. PySpark running on the master VM in your Cloud Dataproc cluster is used to invoke Spark ML functions.


This tutorial uses billable components of Google Cloud Platform, including:

  • Google Compute Engine
  • Google Cloud Dataproc
  • Google BigQuery

Use the Pricing Calculator to generate a cost estimate based on your projected usage. New Cloud Platform users might be eligible for a free trial.

Before you begin

A Cloud Dataproc cluster has the Google BigQuery connector and the Spark components, including Spark ML, pre-installed. To set up a Cloud Dataproc cluster and run the code in this example, you will need to do (or have done) the following:

Linux/Mac OS X


Subsetting data in BigQuery

We’ll use the publicly available natality BigQuery dataset on birthrates to predict birthweight as a function of maternal age, paternal age, and gestation weeks. We use the Google Cloud Client Library for Python to interact with the data.

Run the following code to create a BigQuery table in your default Google Cloud Platform project that contains the selected subset of the natality data.


  1. Run python to open a Python shell on your local machine.
       $ python
       Python 2.7.10 ...
  2. Copy and paste the following code into the Python shell (mouseover the top-right margin in the code block, below, to expose the Click-to-copy widget).
    """Create a Google BigQuery linear regression input table.
    In the code below, the following actions are taken:
      * A new dataset is created "natality_regression."
      * A new table "regression_input" is created to hold the inputs for our linear
      * A query is run against the public dataset,
        bigquery-public-data.samples.natality, selecting only the data of interest
        to the regression, the output of which is stored in the "regression_input"
      * The output table is moved over the wire to the user's default project via
        the built-in BigQuery Connector for Spark that bridges BigQuery and Cloud
    from import bigquery
    from import SchemaField
    from import table
    # Create a new Google BigQuery client using Google Cloud Platform project
    # defaults.
    bigquery_client = bigquery.Client()
    # The name for the new dataset.
    dataset_id = 'natality_regression'
    # Prepare a reference to the new dataset.
    dataset_ref = bigquery_client.dataset(dataset_id)
    dataset = bigquery.Dataset(dataset_ref)
    # Create the new BigQuery dataset.
    dataset = bigquery_client.create_dataset(dataset)
    # In the new BigQuery dataset, create a new table.
    table_ref = dataset.table('regression_input')
    # The table needs a schema before it can be created and accept data.
    # Create an ordered list of the columns using SchemaField objects.
    SCHEMA = []
    SCHEMA.append(SchemaField('weight_pounds', 'float'))
    SCHEMA.append(SchemaField('mother_age', 'integer'))
    SCHEMA.append(SchemaField('father_age', 'integer'))
    SCHEMA.append(SchemaField('gestation_weeks', 'integer'))
    SCHEMA.append(SchemaField('weight_gain_pounds', 'integer'))
    SCHEMA.append(SchemaField('apgar_5min', 'integer'))
    # Assign the schema to the table and create the table in BigQuery.
    table = bigquery.Table(table_ref, schema=SCHEMA)
    table = bigquery_client.create_table(table)
    # Set up a query in Standard SQL.
    # The query selects the fields of interest.
    QUERY = """
    SELECT weight_pounds, mother_age, father_age, gestation_weeks,
    weight_gain_pounds, apgar_5min
    FROM `bigquery-public-data.samples.natality`
    WHERE weight_pounds is not null
    and mother_age is not null and father_age is not null
    and gestation_weeks is not null
    and weight_gain_pounds is not null
    and apgar_5min is not null
    # Configure the query job.
    job_config = bigquery.QueryJobConfig()
    # Set the output table to the table created above.
    dest_dataset_ref = bigquery_client.dataset('natality_regression')
    dest_table_ref = dest_dataset_ref.table('regression_input')
    job_config.destination = dest_table_ref
    # Allow the results table to be overwritten.
    job_config.write_disposition = 'WRITE_TRUNCATE'
    # Use Standard SQL.
    job_config.use_legacy_sql = False
    # Run the query.
    query_job = bigquery_client.query(QUERY, job_config=job_config)

Accessing data in PySpark

Now, we'll load the data from BigQuery and run a linear regression in Python.

  1. SSH into your Cloud Dataproc cluster's master node.

    1. Go to your project's Cloud Dataproc→Clusters page.
    2. Click your cluster name to open the Cluster details page. Then, click the VM instances tab and, at the right on the master instance entry, select SSH→Open in browser window.
    3. A terminal window that is connected to the master instance opens in your browser.
  2. Run pyspark to open a PySpark shell.

  3. Now, copy and paste the following code into the shell to run the Spark ML linear regression (mouseover the top-right margin in the code block, below, to expose the Click-to-copy widget).


"""Run a linear regression using Apache Spark ML.

In the following PySpark (Spark Python API) code, we take the following actions:

  * Load a previously created linear regression (Google BigQuery) input table
    into our Google Cloud Dataproc Spark cluster as an RDD (Resilient
    Distributed Dataset)
  * Transform the RDD into a Spark Dataframe
  * Vectorize the features on which the model will be trained
  * Compute a linear regression using Spark ML


from datetime import datetime
from import Vectors
from import LinearRegression
# The imports, above, allow us to access SparkML features specific to linear
# regression as well as the Vectors types.

# Define a function that collects the features of interest
# (mother_age, father_age, and gestation_weeks) into a vector.
# Package the vector in a tuple containing the label (`weight_pounds`) for that
# row.

def vector_from_inputs(r):
  return (r["weight_pounds"], Vectors.dense(float(r["mother_age"]),

# Use Cloud Dataprocs automatically propagated configurations to get
# the Google Cloud Storage bucket and Google Cloud Platform project for this
# cluster.
bucket = spark._jsc.hadoopConfiguration().get("")
project = spark._jsc.hadoopConfiguration().get("")

# Set an input directory for reading data from Bigquery.
todays_date = datetime.strftime(, "%Y-%m-%d-%H-%M-%S")
input_directory = "gs://{}/tmp/natality-{}".format(bucket, todays_date)

# Set the configuration for importing data from BigQuery.
# Specifically, make sure to set the project ID and bucket for Cloud Dataproc,
# and the project ID, dataset, and table names for BigQuery.

conf = {
    # Input Parameters
    "": project,
    "": bucket,
    "": input_directory,
    "": project,
    "": "natality_regression",
    "": "regression_input",

# Read the data from BigQuery into Spark as an RDD.
table_data = spark.sparkContext.newAPIHadoopRDD(

# Extract the JSON strings from the RDD.
table_json = x: x[1])

# Load the JSON strings as a Spark Dataframe.
natality_data =
# Create a view so that Spark SQL queries can be run against the data.

# As a precaution, run a query in Spark SQL to ensure no NULL values exist.
sql_query = """
from natality
where weight_pounds is not null
and mother_age is not null
and father_age is not null
and gestation_weeks is not null
clean_data = spark.sql(sql_query)

# Create an input DataFrame for Spark ML using the above function.
training_data =["label",

# Construct a new LinearRegression object and fit the training data.
lr = LinearRegression(maxIter=5, regParam=0.2, solver="normal")
model =
# Print the model summary.
print "Coefficients:" + str(model.coefficients)
print "Intercept:" + str(model.intercept)
print "R^2:" + str(model.summary.r2)

The linear regression output (model summary) appears in the terminal window.

<<< # Print the model summary.
... print "Coefficients:" + str(model.coefficients)
<<< print "Intercept:" + str(model.intercept)
<<< print "R^2:" + str(model.summary.r2)
|           residuals|
| -0.7234737533344147|
|  -0.985466980630501|
| -0.6669710598385468|
|  1.4162434829714794|
| 0.32659061654192545|
|  1.5053877697929803|
|  -0.640142797263989|
|   1.229530260294963|
| -0.5160734239126814|
| -1.5165972740062887|
|  1.3269085258245008|
|  1.7604670124710626|
|  1.2348130901905972|
|   2.318660276655887|
|  1.0936947030883175|
|  1.0169768511417363|
| -1.7744915698181583|
only showing top 20 rows

Cleaning up

After you've finished the Use Cloud Dataproc, BigQuery, and Spark ML for Machine Learning tutorial, you can clean up the resources you created on Google Cloud Platform so you won't be billed for them in the future. The following sections describe how to delete or turn off these resources.

Deleting the project

The easiest way to eliminate billing is to delete the project you created for the tutorial.

To delete the project:

  1. In the Cloud Platform Console, go to the Projects page.

    Go to the Projects page

  2. In the project list, select the project you want to delete and click Delete project. After selecting the checkbox next to the project name, click
      Delete project
  3. In the dialog, type the project ID, and then click Shut down to delete the project.

Deleting the Cloud Dataproc cluster

See Delete a cluster.

Deleting Cloud Storage data

To clean up the temporary BigQuery input data files from your project's Cloud Storage bucket:

  1. Run the following commands to obtain the name of the bucket in your project that was used to store the temporary data files (this bucket starts with the characters "dataproc-"). Insert the name of your cluster, as shown below.
    CLUSTER_NAME=<my cluster>
    TEMP_DATA_BUCKET="$(gcloud dataproc clusters describe $CLUSTER_NAME| \
      grep configBucket|awk '{print $2}')"
  2. Go to your project's Cloud Dataproc→Browser page, then click on the TEMP_DATA_BUCKET name. Inside the tmp/ folder, you will see a folder that starts with the name"natality-" (the remaining characters represent the date and time that the folder was created). This is the temporary input data folder to remove. Select the check box to the left of the natality- folder name, then select Delete at the top of the page to remove the folder from Cloud Storage.

Send feedback about...

Google Cloud Dataproc Documentation