- Support for Protobuf version 3.7
- Type hinting
- Serialization of the predict request
- Apply crop when searching by an image
- Reintroduced requests session usage
- Sample milliseconds parameter for video prediction
- Make tag_* methods deprecated
- Rename the embed model used in predict_embed
- Fix setting error object fields on ApiError
- Ignore unknown fields in the response object
- Moderation solution
- Raise ApiError instead of requests.ConnectionError when URL not accessible
- Fix how Protobuf is used to allow simultaneous use of other Protobuf libraries
- Changed the location of the generated proto files to avoid a namespace conflict that appeared in certain cases
- Use Protobuf/gRPC to help with request/response JSON (de)serialization
- Add
assert-code-quality.sh
for static analysis and run it on Travis, fix code style - Add AppVeyor
- Remove unsupported patch outputs and get outputs on inputs
- Use requests.Session to persist and retry connections
- Installing dependencies on Python <3.4
- A dependency version
- More tests
- Made dependency versions ranges instead of using specific versions
- V1 API support (including Pillow dependency)
- Removed checking for upgrades on every client initialization
- Added support for custom face recognition
- Added public models 'enum' object
- Started using semantic versioning
[2.0.32] - PyPI
- workflow predict support
- API_KEY initialization and doc fix
[2.0.31] - PyPI
- retries on all 500 errors
- add a lib dependency
[2.0.30] - PyPI
- expose ModelOutputConfig and ModelOutputInfo
- add search feedback api
- doc fox for feedback api
- refactor feedback api
- fragile test fix
- slash model_id support fix
[2.0.29] - PyPI
- hotfix for base_url regression
[2.0.28] - PyPI
- min_value support
- max_concepts support
- select_concepts support
- feedback support for concept, detection and face detection
- nosetest --with-timer
- code examples for new features
- one-off test code for new features
- doc update and fix from fixit week
- auth key related bug fix
- other minor bugfixs
[2.0.27] - PyPI
- inputs.get_all() mixed status fix
- inputs.get_by_page() mixed status fix
[2.0.26] - PyPI
- verbose in Exception
- less verbose in testing
- strip the url in predict and search
[2.0.25] - PyPI
- bugfix for the api key handling for permission restricted users
[2.0.24] - PyPI
- bugfix for still broken pip3 install
[2.0.23] - PyPI
- bugfix for setup.py for pip3
[2.0.22] - PyPI
- hyper parameters
- patch concepts
- video prediction support
- fix client code bugs
- fix github blocking issue for upgrade
- fix repeatedly predict with Image() bug
- speed up nosetests
- fix multiple concurrent test bugs
[2.0.21] - PyPI
- tutorial for installation on Windows, to address issue #0
- retry on API gateway timeout
- back off on throttled API
- bugfix for
clarifai diagnose
on Windows - other bug and doc fixes
[2.0.20] - PyPI
- propagate the error with more informative text
- expose search score
- bugfix for the token expiration bug
[2.0.19] - PyPI
- get input outputs
- patch input outputs
- shorten the base64 in the test logs
- backoff polling in the sync training
- bugfix on unexpected warnings
[2.0.18] - PyPI
- multi-language support
- test coverage boost
[2.0.14] - PyPI
- python3 support
- add simplified tagging calls
- voluntary resizing in v1 call
- update docs
[2.0.13] - PyPI
- more tests for metadata patch
- doc example for metadata patch
- fix the verbose debug info
[2.0.12] - PyPI
- generalize PATCH for models
- generalize PATCH for inputs
- code refactoring for line width
- more tests
- tutorial for making Image()
- prettier debug info
- remove unused code
- timeout for training
- support filename in Image()
- fix a bug in Python3
- fix Windows compatibility issue