BigQuery DataFrames

BigQuery DataFrames provides a Pythonic DataFrame and machine learning (ML) API powered by the BigQuery engine.

  • bigframes.pandas provides a pandas-compatible API for analytics.

  • bigframes.ml provides a scikit-learn-like API for ML.

BigQuery DataFrames is an open-source package. You can run pip install --upgrade bigframes to install the latest version.

Documentation

Quickstart

Prerequisites

Code sample

Import bigframes.pandas for a pandas-like interface. The read_gbq method accepts either a fully-qualified table ID or a SQL query.

import bigframes.pandas as bpd

bpd.options.bigquery.project = your_gcp_project_id
df1 = bpd.read_gbq("project.dataset.table")
df2 = bpd.read_gbq("SELECT a, b, c, FROM `project.dataset.table`")

Locations

BigQuery DataFrames uses a BigQuery session internally to manage metadata on the service side. This session is tied to a location . BigQuery DataFrames uses the US multi-region as the default location, but you can use session_options.location to set a different location. Every query in a session is executed in the location where the session was created. BigQuery DataFrames auto-populates bf.options.bigquery.location if the user starts with read_gbq/read_gbq_table/read_gbq_query() and specifies a table, either directly or in a SQL statement.

If you want to reset the location of the created DataFrame or Series objects, you can close the session by executing bigframes.pandas.close_session(). After that, you can reuse bigframes.pandas.options.bigquery.location to specify another location.

read_gbq() requires you to specify a location if the dataset you are querying is not in the US multi-region. If you try to read a table from another location, you get a NotFound exception.

Project

If bf.options.bigquery.project is not set, the $GOOGLE_CLOUD_PROJECT environment variable is used, which is set in the notebook runtime serving the BigQuery Studio/Vertex Notebooks.

ML Capabilities

The ML capabilities in BigQuery DataFrames let you preprocess data, and then train models on that data. You can also chain these actions together to create data pipelines.

Preprocess data

Create transformers to prepare data for use in estimators (models) by using the bigframes.ml.preprocessing module and the bigframes.ml.compose module. BigQuery DataFrames offers the following transformations:

  • Use the KBinsDiscretizer class in the bigframes.ml.preprocessing module to bin continuous data into intervals.

  • Use the LabelEncoder class in the bigframes.ml.preprocessing module to normalize the target labels as integer values.

  • Use the MaxAbsScaler class in the bigframes.ml.preprocessing module to scale each feature to the range [-1, 1] by its maximum absolute value.

  • Use the MinMaxScaler class in the bigframes.ml.preprocessing module to standardize features by scaling each feature to the range [0, 1].

  • Use the StandardScaler class in the bigframes.ml.preprocessing module to standardize features by removing the mean and scaling to unit variance.

  • Use the OneHotEncoder class in the bigframes.ml.preprocessing module to transform categorical values into numeric format.

  • Use the ColumnTransformer class in the bigframes.ml.compose module to apply transformers to DataFrames columns.

Train models

Create estimators to train models in BigQuery DataFrames.

Clustering models

Create estimators for clustering models by using the bigframes.ml.cluster module.

  • Use the KMeans class to create K-means clustering models. Use these models for data segmentation. For example, identifying customer segments. K-means is an unsupervised learning technique, so model training doesn’t require labels or split data for training or evaluation.

Decomposition models

Create estimators for decomposition models by using the bigframes.ml.decomposition module.

  • Use the PCA class to create principal component analysis (PCA) models. Use these models for computing principal components and using them to perform a change of basis on the data. This provides dimensionality reduction by projecting each data point onto only the first few principal components to obtain lower-dimensional data while preserving as much of the data’s variation as possible.

Ensemble models

Create estimators for ensemble models by using the bigframes.ml.ensemble module.

  • Use the RandomForestClassifier class to create random forest classifier models. Use these models for constructing multiple learning method decision trees for classification.

  • Use the RandomForestRegressor class to create random forest regression models. Use these models for constructing multiple learning method decision trees for regression.

  • Use the XGBClassifier class to create gradient boosted tree classifier models. Use these models for additively constructing multiple learning method decision trees for classification.

  • Use the XGBRegressor class to create gradient boosted tree regression models. Use these models for additively constructing multiple learning method decision trees for regression.

Forecasting models

Create estimators for forecasting models by using the bigframes.ml.forecasting module.

Imported models

Create estimators for imported models by using the bigframes.ml.imported module.

Linear models

Create estimators for linear models by using the bigframes.ml.linear_model module.

  • Use the LinearRegression class to create linear regression models. Use these models for forecasting. For example, forecasting the sales of an item on a given day.

  • Use the LogisticRegression class to create logistic regression models. Use these models for the classification of two or more possible values such as whether an input is low-value, medium-value, or high-value.

Large language models

Create estimators for LLMs by using the bigframes.ml.llm module.

Create pipelines

Create ML pipelines by using bigframes.ml.pipeline module. Pipelines let you assemble several ML steps to be cross-validated together while setting different parameters. This simplifies your code, and allows you to deploy data preprocessing steps and an estimator together.

  • Use the Pipeline class to create a pipeline of transforms with a final estimator.

ML remote models

Requirements

To use BigQuery DataFrames ML remote models (bigframes.ml.remote or bigframes.ml.llm), you must enable the following APIs:

  • The BigQuery API (bigquery.googleapis.com)

  • The BigQuery Connection API (bigqueryconnection.googleapis.com)

  • The Vertex AI API (aiplatform.googleapis.com)

and you must be granted the following IAM roles in the project:

  • BigQuery Data Editor (roles/bigquery.dataEditor)

  • BigQuery Connection Admin (roles/bigquery.connectionAdmin)

  • Service Account User (roles/iam.serviceAccountUser)

  • Vertex AI User (roles/aiplatform.user)

  • Project IAM Admin (roles/resourcemanager.projectIamAdmin) if using default BigQuery connection, or Browser (roles/browser) if using a pre-configured connection. This requirement can be avoided by setting bigframes.pandas.options.bigquery.skip_bq_connection_check option to True, in which case the connection (default or pre-configured) would be used as-is without any existence or permission check.

ML locations

bigframes.ml supports the same locations as BigQuery ML. BigQuery ML model prediction and other ML functions are supported in all BigQuery regions. Support for model training varies by region. For more information, see BigQuery ML locations.

Data types

BigQuery DataFrames supports the following numpy and pandas dtypes:

  • numpy.dtype("O")

  • pandas.BooleanDtype()

  • pandas.Float64Dtype()

  • pandas.Int64Dtype()

  • pandas.StringDtype(storage="pyarrow")

  • pandas.ArrowDtype(pa.date32())

  • pandas.ArrowDtype(pa.time64("us"))

  • pandas.ArrowDtype(pa.timestamp("us"))

  • pandas.ArrowDtype(pa.timestamp("us", tz="UTC"))

BigQuery DataFrames doesn’t support the following BigQuery data types:

  • ARRAY

  • NUMERIC

  • BIGNUMERIC

  • INTERVAL

  • STRUCT

  • JSON

All other BigQuery data types display as the object type.

Remote functions

BigQuery DataFrames gives you the ability to turn your custom scalar functions into BigQuery remote functions . Creating a remote function in BigQuery DataFrames (See code samples) creates:

  1. A Cloud Functions (2nd gen) function.

  2. A BigQuery connection. If the BigQuery connection is created, the BigQuery service will create a Google Cloud-managed IAM service account and attach it to the connection. You can use a pre-configured BigQuery connection if you prefer, in which case the connection creation is skipped.

  3. A BigQuery remote function that talks to the cloud function (1) using the BigQuery connection (2).

BigQuery connections are created in the same location as the BigQuery DataFrames session, using the name you provide in the custom function definition. To view and manage connections, do the following:

  1. Go to BigQuery in the Google Cloud Console.

  2. Select the project in which you created the remote function.

  3. In the Explorer pane, expand that project and then expand External connections.

BigQuery remote functions are created in the dataset you specify, or in a special type of hidden dataset referred to as an anonymous dataset. To view and manage remote functions created in a user provided dataset, do the following:

  1. Go to BigQuery in the Google Cloud Console.

  2. Select the project in which you created the remote function.

  3. In the Explorer pane, expand that project, expand the dataset in which you created the remote function, and then expand Routines.

To view and manage Cloud Functions functions, use the Functions page and use the project picker to select the project in which you created the function. For easy identification, the names of the functions created by BigQuery DataFrames are prefixed by bigframes.

Requirements

To use BigQuery DataFrames remote functions, you must enable the following APIs:

  • The BigQuery API (bigquery.googleapis.com)

  • The BigQuery Connection API (bigqueryconnection.googleapis.com)

  • The Cloud Functions API (cloudfunctions.googleapis.com)

  • The Cloud Run API (run.googleapis.com)

  • The Artifact Registry API (artifactregistry.googleapis.com)

  • The Cloud Build API (cloudbuild.googleapis.com )

  • The Cloud Resource Manager API (cloudresourcemanager.googleapis.com)

To use BigQuery DataFrames remote functions, you must be granted the following IAM roles in the project:

  • BigQuery Data Editor (roles/bigquery.dataEditor)

  • BigQuery Connection Admin (roles/bigquery.connectionAdmin)

  • Cloud Functions Developer (roles/cloudfunctions.developer)

  • Service Account User (roles/iam.serviceAccountUser)

  • Storage Object Viewer (roles/storage.objectViewer)

  • Project IAM Admin (roles/resourcemanager.projectIamAdmin) if using default BigQuery connection, or Browser (roles/browser) if using a pre-configured connection. This requirement can be avoided by setting bigframes.pandas.options.bigquery.skip_bq_connection_check option to True, in which case the connection (default or pre-configured) would be used as-is without any existence or permission check.

Limitations

  • Remote functions take about 90 seconds to become available when you first create them.

  • Trivial changes in the notebook, such as inserting a new cell or renaming a variable, might cause the remote function to be re-created, even if these changes are unrelated to the remote function code.

  • BigQuery DataFrames does not differentiate any personal data you include in the remote function code. The remote function code is serialized as an opaque box to deploy it as a Cloud Functions function.

  • The Cloud Functions (2nd gen) functions, BigQuery connections, and BigQuery remote functions created by BigQuery DataFrames persist in Google Cloud. If you don’t want to keep these resources, you must delete them separately using an appropriate Cloud Functions or BigQuery interface.

  • A project can have up to 1000 Cloud Functions (2nd gen) functions at a time. See Cloud Functions quotas for all the limits.

Quotas and limits

BigQuery quotas including hardware, software, and network components.

Session termination

Each BigQuery DataFrames DataFrame or Series object is tied to a BigQuery DataFrames session, which is in turn based on a BigQuery session. BigQuery sessions auto-terminate ; when this happens, you can’t use previously created DataFrame or Series objects and must re-create them using a new BigQuery DataFrames session. You can do this by running bigframes.pandas.close_session() and then re-running the BigQuery DataFrames expressions.

Data processing location

BigQuery DataFrames is designed for scale, which it achieves by keeping data and processing on the BigQuery service. However, you can bring data into the memory of your client machine by calling .to_pandas() on a DataFrame or Series object. If you choose to do this, the memory limitation of your client machine applies.

License

BigQuery DataFrames is distributed with the Apache-2.0 license.

It also contains code derived from the following third-party packages:

For details, see the third_party directory.

Contact Us

For further help and provide feedback, you can email us at bigframes-feedback@google.com.