This page documents production updates to BigQuery ML. We recommend that BigQuery ML developers periodically check this list for any new announcements.
You can see the latest product updates for all of Google Cloud on the Google Cloud page, browse and filter all release notes in the Google Cloud Console, or you can programmatically access release notes in BigQuery.
To get the latest product updates delivered to you, add the URL of this page to your
reader, or add the feed URL directly:
November 16, 2021
BigQuery ML is now available in the Santiago (southamerica-west1) region.
September 16, 2021
BigQuery ML documentation has been updated with the following improvements:
- The end-to-end user journey now includes an overview of the machine-learning workflow for each available model.
- Each machine learning module now provides an overview document that describes the BigQuery ML behavior and links to additional guidance. New documentation includes the following:
- Improvements to documentation organization and content, as well as the addition of new landing pages.
August 06, 2021
August 03, 2021
BigQuery ML is now available in the Toronto (northamerica-northeast2) region.
July 28, 2021
July 27, 2021
Explainable artificial intelligence (XAI) helps you understand the results that your predictive machine-learning model generates for classification and regression tasks by defining how each feature in a row of data contributed to the predicted result. This feature is now available for preview.
July 26, 2021
July 19, 2021
The end-to-end user journey for BigQuery ML documents an overview of the complete machine-learning flow for each available model including feature preprocessing, model creation, hyperparameter tuning, inference, evaluation, model export, etc.
June 29, 2021
BigQuery ML is now available in the Delhi (asia-south2) region.
June 22, 2021
BigQuery ML is releasing the following features for preview:
ML.DETECT_ANOMALIESfunction is now available. This function provides anomaly detection for BigQuery ML. The function runs against time-series data using
ARIMA_PLUSmodels. The function runs against independent and identically distributed (IID) random variables data using
AUTOENCODERmodel type is now available for CREATE MODEL statements. This is a TensorFlow-based, deep-learning model that supports sparse data representations, and is commonly used in ML tasks such as feature embedding, unsupervised anomaly detection, and non-linear dimensionality reduction. The ML.PREDICT function can use previously built AUTOENCODER models to reduce the dimensionality of query results.
- Hyperparameter tuning is now available and can be used to improve model performance by searching for the optimal hyperparameters when training ML models using CREATE MODEL statements. View the BigQuery ML Hypertuning tutorial to learn how to improve model performance by 40%.
June 21, 2021
BigQuery ML is now available in the Melbourne (australia-southeast2) region.
May 18, 2021
CREATE MODEL statement for training AutoML Tables models is now generally available (GA). AutoML Tables enable you to automatically build state-of-the-art machine learning models on structured data at massively increased speed and scale. For more information, see
CREATE MODEL statement for training AutoML Tables models.
April 19, 2021
BigQuery ML is introducing new ARIMA_PLUS models and deprecating the ARIMA model type. While the underlying modeling technique has not changed, the following improvements are now available in ARIMA_PLUS:
Multiple ID columns are specifiable via
Additional time series (500,000) for simultaneous forecasting.
Finer data frequency:
March 24, 2021
BigQuery ML is now available in the Warsaw (europe-central2) region.
March 11, 2021
BigQuery ML now supports training for DNN/Boosted Tree models in the Iowa (us-central1) region.
January 19, 2021
BigQuery ML is now available in the Iowa (us-central1) region.
November 23, 2020
BigQuery ML integration with AI Platform for Deep Neural Network (DNN) models is now generally available (GA). For more information, see
CREATE MODEL statement for Deep Neural Network (DNN) models.
September 29, 2020
Time series models now let you change
DATA_FREQUENCY from the default value (
AUTO_FREQUENCY) when forecasting multiple time series using
August 27, 2020
For more information about time series model support, see the following documentation:
August 17, 2020
Matrix Factorization model support is now Generally Available (GA). For more information, see the following documentation:
August 06, 2020
BigQuery ML is now available following regions: Oregon (us-west1), Belgium (europe-west1), and Netherlands (europe-west4).
July 15, 2020
Data split and validation options are now available for AutoML Table model training.
July 01, 2020
June 16, 2020
AutoML Tables models. For more information, see CREATE MODEL statement for AutoML Tables models.
Boosted Tree models using XGBoost. For more information, see CREATE MODEL statement for Boosted Tree models.
Deep Neural Network (DNN) models. For more information, see CREATE MODEL statement for DNN models.
June 08, 2020
BigQuery ML is now available in the Jakarta (asia-southeast2) region.
April 27, 2020
BigQuery ML is now available in the Las Vegas (us-west4) region.
April 22, 2020
April 17, 2020
BigQuery ML now supports Matrix Factorization models for recommendations, as a beta release. For more information, see The CREATE MODEL statement for Matrix Factorization.
February 24, 2020
BigQuery ML is now available in the Salt Lake City (us-west3) region.
January 24, 2020
BigQuery ML is now available in the Seoul (asia-northeast3) region.
December 19, 2019
BigQuery ML data preprocessing is now Generally Available (GA). Read about the preprocessing functions and walk through how to use the TRANSFORM clause for feature engineering.
December 04, 2019
You can now use KMEANS++ to initialize the clusters of a k-means model. KMEANS++ trains a better model than random cluster initialization.
November 21, 2019
BigQuery ML data preprocessing is now in beta.
BigQuery ML now supports customer-managed encryption keys (CMEK). You can use your own Cloud KMS keys to encrypt ML models.
November 20, 2019
BigQuery ML is now available in the South Carolina (us-east1) region.
September 30, 2019
September 23, 2019
September 17, 2019
BigQuery ML is now available in the Frankfurt (europe-west3) region.
July 02, 2019
May 29, 2019
May 14, 2019
BigQuery ML now supports the DROP MODEL DDL statement for deleting models.
May 06, 2019
BigQuery ML IAM permissions are now available. These permissions take effect on June 6, 2019. Customers with custom roles should migrate to these permissions no later than June 6. Pre-defined IAM roles and basic roles are not impacted by this change.
April 10, 2019
BigQuery ML now supports the k-means model type for clustering and customer segmentation.
April 04, 2019
During the beta period, Table permissions were automatically applied to models for custom IAM roles. BigQuery ML will begin enforcing several new IAM permissions on June 6, 2019. Customers who used custom IAM roles during the beta period must reconfigure these roles to use the new BigQuery ML permissions. This change will enable you to manage Models permissions separately from BigQuery ML Table permissions. You can begin redefining your custom roles by the end of April, 2019 when the permissions are released. Pre-defined IAM roles and basic roles are not impacted by this change.
March 18, 2019
The limit on the number of
CREATE MODEL queries has increased from 100 to 1,000.
January 29, 2019
BigQuery ML now supports automatic, batch gradient descent, and normal equation optimization strategies for linear regression models.
December 13, 2018
The BigQuery ML
ML.WEIGHTS function now supports standardization.
The BigQuery ML
ML.PREDICT function now supports thresholds for binary logistic regression models.
November 08, 2018
BigQuery ML pricing is now available.
October 19, 2018
The BigQuery ML
CREATE MODEL statement has increased support for unique values in labels from 10 to 50. Multiclass logistic regression models now support up to 50 unique values for labels.
October 11, 2018
When you create a model using the
random data split method, the split is now deterministic. Subsequent training runs will produce the same split so long as the underlying input data hasn't changed.
Providing input data to the
ML.EVALUATE function is now optional.
September 19, 2018
BigQuery ML is now available in the Tokyo (asia-northeast1) region.