TensorFlow in versions >=2.1<=2.3 stated SciPy as a dependency but it is not used in the codebase

Even though SciPy is stated in the requirements of TensorFlow>=2.1<=2.3, it is not used in TensorFlow’s codebase.

Issue description

TensorFlow releases matching version specification >=2.1<=2.3 stated SciPy package as a dependency. This dependency is used only in TensorFlow testsuite and is not used when running TensorFlow as a library.

The installation process time can be affected negatively as SciPy can trigger building native extensions. Moreover, shipping the application with code that is not used has additional disadvantages.

Affected packages

tensorflow>=2.1<=2.3
tensorflow-cpu>=2.1<=2.3
tensorflow-gpu>=2.1<=2.3
intel-tensorflow>=2.1<=2.3
scipy==1.4.1
scipy==1.2.2

Severity

WARNING

Issue fix

Resolver will automatically remove SciPy and all its dependencies from the application dependencies if the application does not use this package (directly or indirectly using other dependencies).

Pipeline units

TensorFlowRemoveSciPyStep

Recommendation types

All the recommendation types can produce this message, except for latest:

performance
security
stable
testing

See this document that describes recommendation types listed.