Skip to content

Release notes

Test releases (0.1.x) are used to test the package, the installation and the Microsoft Marketplace process on Databricks. They are not offered to customers. The first public release is 0.2.0.

DoubleML PySpark 0.2.0 (planned)

First public release, available in the Microsoft Marketplace. Contents of 0.1.3 plus the changes from the final tests.

DoubleML PySpark 0.1.3 (test release)

Test release of the complete Marketplace setup.

Platforms

  • Databricks serverless compute (environment versions 4, 5 and 6), Databricks Runtime 17.3 LTS, 18 LTS and 19
  • Databricks Runtime 16.4 LTS (Spark 3.5) on clusters in dedicated access mode
  • Install from the Marketplace package index with your license key, with pip or uv (see Install from the Microsoft Marketplace)

Models and results

  • Partially linear regression (PLR) and interactive regression model (IRM)
  • Heterogeneous effects: CATE and GATE with pointwise and joint confidence intervals and robust covariance (HC0 to HC3, default HC0)
  • Hyperparameter tuning with Optuna, sensitivity analysis for unobserved confounding
  • summary, coef, se, all_coef, all_se and pval as in DoubleML for Python

Other changes

  • The example data generators need the optional extra datasets: pip install "doubleml-pyspark[datasets]"
  • The license check also sends the Databricks workspace and a pseudonymised user identifier, to count usage per license (see Privacy notice)

DoubleML PySpark 0.1.2rc1 (test release, 16 October 2026)

First compiled package (wheel) for Databricks serverless compute and clusters, installed from the package index. Tested on serverless compute and Databricks Runtime 18 LTS.

DoubleML PySpark 0.1.1 (internal, 8 October 2026)

Container image for Databricks clusters. Replaced by the wheel, which also runs on serverless compute.