Skip to content

Latest commit

 

History

History
63 lines (46 loc) · 2.45 KB

README.md

File metadata and controls

63 lines (46 loc) · 2.45 KB
Raw

TFX package

TFX is packaged as the tfx package on PyPI. We recommend that users install TFX using pip install tfx. As of version 0.26.0, users also have the option to install a standalone version of the TFX pipeline authoring SDK, as the ml-pipelines-sdk package. This package has minimal dependencies, but does not include first-party TFX components like the TFX ExampleGen, Transform and Trainer in tfx.components.*, nor any additional tools requiring these components.

Both the tfx and ml-pipelines-sdk packages share the tfx namespace and the same source repository at https://github.com/tensorflow/tfx. These two packages can be built using the instructions below. During development, a single editable package may be installed for convenience (see the "Installing the development-only tfx-dev package" section below).

Building TFX pip package from source

Setting up the build environment

First, set up the build environment by running:

package_build/initialize.sh

Building the tfx and ml-pipelines-sdk packages

Next, each package can be built using the bdist_wheel command:

python package_build/ml-pipelines-sdk/setup.py bdist_wheel
python package_build/tfx/setup.py bdist_wheel

As a result, .whl files will be generated in the dist/ directory.

Installing the development-only tfx-dev package

During development, it is convenient to install a single editable pip package. This package will contain the union of the tfx and ml-pipelines-sdk package in an editable environment. To install this combined package for development, run from the repository root:

pip install -e .

This tfx-dev package should not be packaged as a binary or source distribution using python setup.py {bdist_wheel,sdist} to avoid conflicts with the two official tfx and ml-pipelines-sdk packages. Instead, users should build the two packages for distribution with the directions above.

Temporary workaround for building tfx-dev wheels.

To minimize dependency issues, the instructions above should be used to build TFX wheel files for deployment. As a temporary workaround, the environmental variable UNSUPPORTED_BUILD_TFX_DEV_WHEEL may be set to 1 to forcibly enable building and installation of a single tfx-dev pip package containing the union of the tfx and ml-pipelines-sdk packages. This workaround may lead to package namespace conflicts and is not recommended or supported, and will be removed in a future version.