* Rename docs directory
* Add documentation dependencies
* Add documentation dependencies to `setup.py`
* Add javascripts
* Add CSS stylesheets
* Add TF logo
* Add readme as landing page
* Add basic mkdocs config
* Use correct index page
* Add images for index page
* Add install and getting started pages to navigation
* Add API docs
* Add docs workflow
* Remove deprecated code
* Check links with lychee in docs workflow
* Revert "Check links with lychee in docs workflow"
This reverts commit 21b34867eb.
* Add anomalies reference
* Fix bad links in docs
* Fix links that should be internal
* Run pre-commit
* Add build workflow via docker (#259)
* Add build workflow via docker
* rename docker-compose to docker compose
* add twine check and upload to PyPi
* add workflow_dispatch
* install twine before twine check
* add testing workflow
* single python
* trigger
* install in build job
* install pytest
* install test dependencies
* add xfail to tests
* add reusable workflows and add pr number in xfail
* fix composite action
* add more xfails
* xfail top_k_uniques_stats_generator_test.py
* xfails in partitioned_stats_generator_test.py
* more xfails
* add missing imports
* fix extra decorators
* more xfails
* Fix TAP and Kokoro tests caused by NumPy v2 migration.
1. To ensure test compatibility between NumPy v1 and v2 environments, we've adjusted the comparison tolerance to 1e-4. This accommodates slight variations (around 1e-4) in floating-point outcomes between the two NumPy versions. Additionally, we've modified the expected proto float to align with NumPy v2 results.
2. For mutual_information, NumPy v2 is able to handle values > 2**53 if the min and max of the examples are the same. However, since we need to be compatible with NumPy v1 and v2, for related unit tests, we check for the NumPy version before running the associated unit tests.
PiperOrigin-RevId: 681598675
* use xfail instead of skip
* remove xfails that are passing
* dont run xfail + add test deps
* fix build failure by pinning tensorflow_metadata
* move test requirements
* debugging
* more debugging
* remove upload for testing
* add environment variable to build nightly
* add extra-index-url
* trying to use nightly install
* revert debugging changes
* update upload artifact version
* revert metadata branch back to master
* fix typo
* remove install when built, move to only install on test
* change name of step checking the wheel after moving install to test workflow
* update PR number
* just remove PR
---------
Co-authored-by: Amit Kumar <dtu.amit@gmail.com>
Co-authored-by: tf-data-validation-team <tensorflow-extended-nonhuman@googlegroups.com>
Improvements to Beam's type hinting infrastructure found a breakage in this code based on mismatched type hints (stemming from the one-way relationships between lists, sequences, and iterables.) CoGroupByKey outputs Iterables, not Sequences, but these type checks were errantly passing before.
PiperOrigin-RevId: 733394393
Protobuf 6.30.0 will change the return types of Descriptor::name() and other methods to absl::string_view. This makes the code work both before and after such a change.
PiperOrigin-RevId: 689135228
* add testing workflow
* single python
* trigger
* install in build job
* install pytest
* install test dependencies
* add xfail to tests
* add reusable workflows and add pr number in xfail
* fix composite action
* add more xfails
* xfail top_k_uniques_stats_generator_test.py
* xfails in partitioned_stats_generator_test.py
* more xfails
* add missing imports
* fix extra decorators
* more xfails
* use xfail instead of skip
* remove xfails that are passing
* dont run xfail + add test deps
1. To ensure test compatibility between NumPy v1 and v2 environments, we've adjusted the comparison tolerance to 1e-4. This accommodates slight variations (around 1e-4) in floating-point outcomes between the two NumPy versions. Additionally, we've modified the expected proto float to align with NumPy v2 results.
2. For mutual_information, NumPy v2 is able to handle values > 2**53 if the min and max of the examples are the same. However, since we need to be compatible with NumPy v1 and v2, for related unit tests, we check for the NumPy version before running the associated unit tests.
PiperOrigin-RevId: 681598675
William Black