CHANGES
=======

* Fix #71, remove non-ascii characters from README
* Revert "try removing description file"
* try removing description file
* Add utf8 encoding flag for setup.cfg

0.7.3
-----

* Release 0.7.3
* Pin lib version in ephemeral TFDV setup (#217)
* Fix typo in supported dataflow args list (#218)
* Add docs for status-options, ignore .vscode/ (#216)

0.7.2
-----

* Release 0.7.2
* Fix #213, #207 - Bump Tensorflow version: 1.13.1 -> 1.14.0 and tfx libs to 0.14.x (#208)

0.7.1
-----

* Release 0.7.1
* Use model\_name\_suffix in \_get\_job\_name() (#209)
* Add python-version argument to tensorflow-task (#210)
* Fix python3 dataflow naming (#214)
* Drop required-python line from setup.cfg (#212)
* Update verify-license-headers (#205)
* Update LICENSE copyright holder (fixes #201) (#202)
* tfdv: expose infer\_feature\_shape flag (#200)

0.7.0
-----

* Release 0.7.0
* Remove default output handling in tensorflow task (#181)
* rename gcloud commands (#197)
* Use tfx dependency (#196)

0.6.7
-----

* Release 0.6.7
* Pin scikit-learn to version supporting Py2 (#195)

0.6.6
-----

* Release 0.6.6
* Expose validate\_statistics parameters (#189)
* TFDV: Allow compressed inputs (e.g. tfrecord.gz) (#191)

0.6.5
-----

* Release 0.6.5
* Fix #186, allow optional schema inference. (#188)
* Python 3 support (fixes #176) (#179)
* Fix #184 allow stats options to be passed to TfDataValidator
* push default job name handling up to python dataflow task
* Fix #178 remove tf versions from envlist

0.6.4
-----

* Release 0.6.4
* Bump tfx libs to 0.12
* Bump to tensorflow 1.13.1
* Bump tensorflow version to 1.13.0rc1
* Expose \_get\_job\_name in TensorFlowTask
* fix spacing change
* enables using binary schema files for tfdv

0.6.3
-----

* Release 0.6.3
* Update setup.py content
* Clean up README and example
* Remove reference to tfdv extras in readme
* Combine GenerateStats and TfDataValidator (#165)
* Consolidate tfdv.GenerateStats (#164)
* update readme

0.6.2
-----

* Release 0.6.2
* Add more release instructions to readme
* Add timestamp to job name to make unique (#162)
* rm print statements

0.6.1
-----

* Fix output handling in tfdv task (#160)
* rename tfx extras to tfdv (#158)
* Move luigi to extras (#157)
* Use tf 1.9.0 for everything (#156)
* lint: check license on every py file (fixes #151) (#154)
* Fix output publishing in GenerateStatsTask
* tfr\_read.py: accidentally has executable permission (#153)
* tests: run against TF 1.11+ w/o segfault (#150)

0.6.0
-----

* Release v0.6.0
* Move tensorflow-transform to requirements.txt (#149)
* Remove auto generated tft requirements file (#146)
* Code clean up
* Add release cmd to readme (#144)
* Add requests in requirements.txt (#143)
* Remove test-requirements (#141)
* Fix #136 use schema\_utils from TFT (#142)
* reqs: luigi {tfx => core} (#140)
* extras: drop tensorflow-gpu (#137)
* tfr-read: simplify impl and add a smoke test (#127)
* extras: drop tensorflow extra (move to reqs.txt) (#138)
* Rename \`pipeline\` to \`tfx\` (#133)
* Make preprocessing\_fn a memeber in TFTransform
* Add tf.transform example code
* Clean up code
* Also add temp\_location to pipeline\_args
* Add unit tests for tft
* Allow to use pre-saved transform\_fn on evaluation data
* Add tft luigi task
* Add tf.transform support
* tests: run nosetest in verbose mode (#132)
* build: coverage was broken by 8475791 (#131)
* build: simplify, add extras\_require, tf variants (#125)
* tfr-read: move to main code tree (#126)
* Bring .mailmap up-to-date (fixes #93) (#121)
* README: Update coverage badge to be unbranded (#118)
* Cleanup warnings, errant files/dirs, mypy for tests (#107)
* tests: Increase coverage for TensorflowTask (#115)
* tests: More tests for spotify\_tensorflow.luigi.utils (#117)
* tests: Add tests for spotify\_tensorflow.luigi.bq module (#116)
* Add DeprecationWarning for ExternalDailySnapshot (#114)
* luigi.tfdv: \_stats.pb placed in same dir as data (#108)
* tensorflow\_task: use run\_with\_logging (#112)
* tensorflow\_task: dedupe get\_uri logic (#111)
* Update CircleCI py3 image 3.6.{1->7} (#110)
* Update CircleCI py2 image 2.7.{14->15} (#109)
* Add TFRecordsStatsTask using TFXBaseTask (#103)
* Add tfx luigi base task
* Add python dataflow base luigi task
* add input args management back in
* add new beam base task
* Revert "Add TFDV worker harness image (with build) (#90)" (#101)
* Add TFDV worker harness image (with build) (#90)
* Add Datasets.parse\_schema\_from\_stats fn (#89)
* Remove setup.cfg ref to non-existent module (#88)
* Remove tensorflow\_transform monkeypatching (#87)
* tox: downscope flake8 (was running over local venv) (#86)
* #55: Incorrect job\_dir param description (#84)
* Tensorflow: 1.9.0 => 1.10.1 (#82)
* README: update examples section (#81)

0.5.1
-----

* Release v0.5.1
* Add util to re-order datasets based on featran settings (#79)

0.5.0
-----

* Release v0.5.0
* remove pinning of dependencies on patches
* add spase unit test
* refactor test
* add documentation for default\_value
* add default values to dict and dataframe APIs
* Use seperate methods
* TF Record Spec is no more

0.4.0
-----

* Release v0.4.0
* Fixes: #72, #57, #42, #24, #62. Remove the use of tf.flags. Breaking changes
* setup\_requires pbr in setup.py
* Depend on setuptools due to pbr
* Cosmetic changes
* Depend on tensorflow\_metadata
* Remove tox integration from setup.py
* Fix #49. Remove trainer, use std estimator api
* implement dir globing in tfr-read. fix #66
* fix IOError.EPIPE bug
* Bump TF to 1.9.0
* changes based on PR comments
* fix python3 build
* fail build if schema.proto changed
* checkin compiled schema proto to workaround lack of python2 support in tensorflow\_metadata
* checkin bits of tensorflow\_transform to work around lack of py3 support
* fix style
* delete old decoding code
* add tests and fix some bugs
* add headers
* create command line tool for reading tfr records as json
* refactor code
* wip, command line tool for reading tfrecords
* use iris for examples
* Add runtime version and scale tier params (#60)

0.3.2
-----

* Release 0.3.2
* Make featran dataset method public. (#56)

0.3.1
-----

* Release v0.3.1
* Type as much as possible - fix bugs on the way
* Fix link in doc
* Skip some other files from dist
* Don't include scripts in the dist

0.3.0
-----

* Release v0.3.0
* update luigi task to pass 'job-dir'
* Add stuff to readme
* migrate existing code to absl flags
* add flag testing
* rename flags

0.2.15
------

* Release v0.2.15
* Support freezing to GCS + typing of FreezeGraph
* Discourage freezing graph - for now just doc
* Bump tf to 1.8.0
* Tox up

0.2.14
------

* Release 0.2.14
* fix job-dir arg

0.2.13
------

* Release v0.2.13
* make model\_package and model\_package\_path non optional parameters
* fix examples imports
* fix tests
* add model\_package\_path to luigi
* add luigi param description
* tensorflow task doesn't force requires override
* Revert "Define a bit of public API"
* Fixup circleci setup
* Add examples - tf and sklearn

0.2.12
------

* Release v0.2.12
* Upgrade TF to 1.7.0
* Upgrade TF to 1.6.0

0.2.11
------

* Release v0.2.11
* Add get\_context
* Define a bit of public API
* Remove obsolete print
* Clean up dummy readme

0.2.10
------

* Release v0.2.10

0.2.9
-----

* Release v0.2.9
* vstack instead of hstack
* Fix #23. Upgrade to tf 1.5.0
* Mailmap for rich
* Dictionary endpoint for spotify-tensorflow
* Read Featran Settings file and also return feature names
* Method to freeze a TF graph (#26)
* Apply fix from spotify/spotify-tensorflow/pull/11
* Release v0.2.9
* fix order problem
* add code to generate tf test resources

0.2.8
-----

* Release v0.2.8
* Remove author - this is auto generated anyway
* add test for trainer.py
* fix trainer bugs
* rm as\_np helper argument
* add Pandas DataFrame endpoint
* changes based on review
* add integration test
* support TFRecordSpec
* change based on comments
* rm outdated example
* cleanup dataset class
* use namedtuple
* use pep8 order for imports
* refactor dataset iterator logic
* add records prefetching
* minor flags refactoring
* Add unit tests for tensorflow task. (#10)

0.2.7
-----

* Release v0.2.7
* add sharding for distributed execution (#9)
* Update flags.py

0.2.6
-----

* Release v0.2.6
* Fix job name of cloud mode
* Refactor tensorflow\_task

0.2.5
-----

* Release v0.2.5
* Bring back the side effect of making a job name

0.2.4
-----

* Release v0.2.4
* Bring back default RunConfig helper (this time in Trainer)
* Refactor Experiment handling
* Add doc for Experiment
* Refactor get\_featran\_example\_dataset, make it more readable
* Fix import order in bq
* Move from namedtuple to proper class, add doc
* Parametrize feature mapping function, add more docs
* Fix pythondoc, add more info, use google style
* Fix import of ExternalDailySnapshot
* Refactor Experiment - parametrize, simplify and document
* Simplify and parametrize RunConfig in Trainer
* Refactor flags
* Refactor trainer API
* Fix experiment - use combined dataset API
* Remove obsolete featran helpers
* Enforce import order - follow google style
* Configure sdist in more sensible way

0.2.3
-----

* Release v0.2.3
* It's TensorFlow not Tensorflow
* Some small typos/style fixes
* Exclude vevn used in circleci testing
* Accept only utf-8 coding
* Fix and enforce coding
* Configure flake8 excludes
* Fix py2,3 compatibility
* Remove obsolete comment - already fixed
* Fix and enforce pystyle
* fix flags parsing
* add take\_count arg to dataset
* add luigi utils to lib
* more reasonable default value for buffer\_size

0.2.2
-----

* Release v.0.2.2
* relative imports for python3
* relative import for both python2/3
* change floats to int and set compression type
* change test to match Datasets return
* update to latest from working example

0.2.1
-----

* Release v0.2.1
* Add tensorflow dep to tests
* Move TF dependency to extras

0.2.0
-----

* Release v0.2.0
* Update to TF 1.4.0

0.2.0.rc1
---------

* Fix py3 style
* Fix style
* Get rid of gen\_spec, add feature names to the context
* Upgrade to tf 1.4.0rc1
* Use seperate cache file for py2
* Just test really
* Use virtualenv
* Use virtualenv for py2
* Add workflow
* Add python2 circleci test
* Add badge for codecov
* Forgot to active the venv for the codecov
* Use same style as circleci
* Add codecov
* Update readme
* Add circleci config

0.1.1
-----

* Publish also for python 2
* Add example of use of featran dataset
* Return actual number of features
* Add compression to example dataset

0.1.0
-----

* Update ignore files
* Requires >= on TF
* Update ignore files
* Remove obsolete readme parts
* Remove spotify build files
* Add license and notice
* Add some setup cfg config
* Fix style
* Fix style add nose dep
* Ignore swp files
* Add test for featran dataset
* Empty list instead of None
* Use TF 1.3.0
* Fix style
* Add test requirements
* Remove default value for features
* Expose parsing function generator
* Add requirements.txt
* Fix indentation, from 2 to 4
* Use file\_io
* Dataset Example reader helper
* Init commit
