v1.1.2

SDK-only release compatible with Tune Insight v1.1.*. Install with

pip install tuneinsight==1.1.2

This version marks some dependencies (mostly related to jupyter) as optional. The above will not include these dependencies and is recommended for production environments. Users who intend to use the SDK in notebooks should do the full install:

pip install tuneinsight[full]==1.1.2

(If you are using zsh, you may need to escape the brackets as tuneinsight\[full\]).

This release also updates some of the dependencies of the SDK, in particular python-keycloak. It also includes upcoming features (of v1.2) that might not be supported by your instance, but should be backwards compatible.

v1.1.0

Compatible with Tune Insight v1.1.0. Install with

pip install tuneinsight==1.1.0

This version mirrors the major version change for the Tune Insight solution. Changes to the SDK are minor. New features include:

• Better usability by non-editor users (a bug previously made such users unable to run projects).
• Ability to check what a user is allowed to do (capabilities) from the SDK with client.can.
• Projects can now be (un)archived.
• Units can be specified to filter records such that they use the appropriate units for a numerical value.

v1.0.1

Compatible with Tune Insight v1.0.1. Install with

pip install tuneinsight==1.0.1

This is a minor patch version that mostly includes UI changes, and this SDK release is intended for consistency. It is (99%) compatible with v1.0.0., but includes additional features supported by the API but intended for the SDK.

v1.0.0

Compatible with Tune Insight v1.0.0. Install with

pip install tuneinsight==1.0.0

This version includes major (mostly backwards-compatible) changes, conforming to the changes in the Tune Insight solution to v1.0.0. New features available in the solution and mirrored in the SDK include:

• The Feasibility, Distribution and FilteredAggregation computations, which all extend aggregations for specialized use cases.
• Automated dataset schema inference.
• Policies at a datasource level.
• Differentially private statistics.
• Confidence intervals and post-processing for differentially private survival curves.
• Screening operations for datasets.
• Privacy budgets can be scoped at a datasource level.

The following new features were added to the SDK specifically:

• Support for remote datasources (for non-contributing nodes).
• A human-readable display of available project actions.

This release also includes many smaller changes aiming to improve the user experience, such as improved error messages.

v0.14.0

Compatible with Tune Insight v0.14.0.

As for v0.13.0, changes in the SDK are minor, mostly focusing on compatibility with changes in the backend.

• Support has been added for E2EE statistics. This means that the cryptolib has changed.
• Additional packages were erroneously included in the cryptolib, increasing its size by about 25%. This has been fixed.
• Local data selections were deprecated in the backend. However, this concept is still useful in the SDK, and has thus been kept (with a different underlying implementation).

v0.13.0

Compatible with Tune Insight v0.13.0.

Changes to the SDK are mostly minor with this version focusing on backend and web interface.

• The preprocessing has been improved, as well as the RemoteDataFrame asbtraction.
• Datasources can now be defined at the datasource level.
• Deprecation of all project.new_{computation} methods -- use the tuneinsight.computations.Computation classes instead.

v0.12.1

Compatible with Tune Insight v0.12.1.

This release includes mostly minor changes on the SDK side:

• Quality-of-life improvements to make the SDK easier to use, including context managers to define policies (with project.policy as policy: ... instead of get_policy then set_policy)
• Improvements to docstrings and documentation.
• Changes to the EncryptedMean computation interface.
• Updates to the cryptolib for compatibility with Lattigo v6.5

v0.11.1

Compatible with Tune Insight v0.11.1.

Major changes

• The authorization mechanism has been strengthened: all projects must now be authorized and are locked upon authorization request.
• Computations can now be fetched from a project (creating the appropriate Python object). This can be used to apply consistent post-processing of results computed in other sessions.
• The cryptolib has been been tested and works on Linux, Windows, and Mac OS.

Minor changes

• The regression has been improved and should be easier to use.
• Synthetic data generation has been improved (including with the addition of a new progress tracking API).
• Features marked for deprecation in this release have been deprecated. The next deprecation wave should come with v0.12.x.
• Many minor features have been added to reflect changes in Tune Insight v0.11.1 (e.g., recurring projects, joining projects with tokens, ...)

v0.10.3

Compatible with Tune Insight v0.10.3

Major changes

• Retrieval of computations and results from the project.
• EncryptedAggregation and StatisticalAggregation merged as one computation.

Minor changes

• Various quality-of-life improvements.
• Removal of experimental/outdated features.

Release package on the docs: https://docs.tuneinsight.com/packages/tuneinsight-diapason-0.10.3.tar.gz

v0.10.2

Compatible with Tune Insight v0.10.2

Release package: https://docs.tuneinsight.com/packages/tuneinsight-diapason-0.10.2.tar.gz

Full Changelog: v0.9.2...v0.10.2