Releases: France-Travail/gabarit
Gabarit 1.3.3
Improvements:
PR167 : Added the random_seed parameter for the models of the NLP and NUM templates
PR170/PR171 : Updated pydantic to v2 in order to be able to use fastapi>=0.100
Bugfixes :
PR167 : Fixing the random seed closes the issue #60 on a non determinist test
PR168 : Fixed pillow version to 9.5.0 to avoid a bug with detectron
PR172 : Updated the version of pycocotools to avoid a bug where it couldn't be build
PR174 : Fixed memory leaks in alternative predicts
PR176 : Updated nanasess/setup-chromedriver to v2
Gabarit 1.3.2
Improvements:
PR159 : Renamed the Dockerfile.svc to Dockerfile in the template API
Bugfixes :
PR155 : Fixed a bug in model_keras and model_huggingface where there was an equality test instead of an assignement (and created the corresponding unit tests which were missing)
PR159/PR162 : Fixed a bug in ModelKeras where the experimental version of predict created a serve function multiple times. Also changed the name of the experimental version to alternative_version (and created corresponding unit tests)
PR163 : Fixed a bug in the documentation generation which prevented the documentation of the template API to be properly generated
Gabarit 1.3.1
Improvements:
PR145 : Instantiation of models in the template API now uses lifespan events
PR149 : All the assert statements (outside of tests) have been removed
Bugfixes:
PR149 : Corrected a bug in the API template tests where a np.float128 could make the test crash on windows
PR149 : The NAME_VENV was never used in the makefile of the API template
PR150 : Corrected a bug in the sweetviz folder path in the numeric tutorial
Gabarit 1.3.0
New:
PR46 : Added tutorial for numerical template
PR54 : Added tutorial for vision template
PR69 : Added Sweetviz in the numeric and NLP templates in order to produce EDA reports
PR82 : Added an explainer to the vision template and improved the explainer for the NLP template
PR83 : Added fairness reports in the numeric template
PR88 : Added an explainer to the numeric template and fixed some bugs for the existing explainers
PR89 : Added an aggregator of models to create a meta model in the numeric and NLP templates
PR94 : Added the template API to gabarit
PR99 : Removed pytorch lighntning and added a ModelHuggingFace to use HuggingFace models
PR122, PR128 and PR129 : Added mkdocs documentation
Improvements:
PR44 : Deleted the with_new_embedding argument
PR58 : Removed all references to nb_iter_keras
PR59 : Mutualization of predict_proba in model_keras, check for the input format in the classifier models of the numeric template
PR45 : Improved the required packages versions
PR51 : Updated tensorflow requirements for the NLP and numeric Templates
PR65 : Added a warning when using custom functions in the preprocessing of the numeric template
PR73 : Refactoring of the use of MLflow in gabarit
PR62 : Replaced get_ct_feature_names by get_feature_names_out from sklearn.compose.ColumnTransformer
PR96 : Fixed some typos in the template API
PR102 : Added HuggingFace proxies option
PR112 : Added some unit tests for the vision template
PR107 : Uniformize the way utils_models.predict works between the templates
PR125 : Add gabarit version in the configurations.json and properties.json
PR103 : Changed the hierarchy of the model folders to be more organized
PR109 : Added a warning when using detectron2 on windows
PR110 : Removed unused command in the makefiles
PR106 : Added the --overwrite argument to some script
PR130 : Pickle has been replaced by dill everywhere
PR131 : Added sentencepiece as a dependency in the NLP template
Deleted the mypy requirements in setup.py
Updated .gitignore
Updated readme with a procedure to update a project with a new gabarit version
PR138 : Changed the way to reload a model (internally)
PR139 : The package can't contain hyphens "-" in their name. They are automatically replaced by underscores "_"
Bugfixes:
PR95 : Fixed two small bugs in the fairness reports script
PR105 : Fixed a bug where Streamlit was not working with the ModelExplainer based on Lime
PR111 : Fixed a bug where MLflow artifacts did not appear in the MLflow UI
PR114 : Fixed a bug in the template API where the type annotations prevented some unit tests to pass
PR120 : Fixed a bug concerning the artifact_uri in MLflow
PR132 : Added add_model_length attribute in the ModelHuggingFace (to be able to use camembert/camembert-base)
Fixed the pathspec requirements to avoid a bug with DVC
PR138 : The local version of a gabarit package now installs with a "+local" instead of a "-local"
Gabarit 1.2.4
Bug fix :
PR 43 : Gabarit (fix): fixed an issue with uneeded dependancy to psutil
Fix github url in setup.py
Gabarit 1.2.3
Improvements :
- PR 17 : Various minor changes
- Improved gitignore
- Various type hint updates
- Various pep8 updates
Bug fix :
- PR 14 & PR 17 : Minor fix for NLP tutorial
- PR 18 : Fixed rare cases where validation data was missing some classifiers targets, resulting in bad y shape
- PR 42 : Temporary fix for scipy version (<1.9)
Gabarit 1.2.2
New:
PR8 & PR13: Added a new NLP tutorial + a French NLP tutorial
Improvements:
PR6: Some more translations (French -> English)
PR7: Many linter fixes
PR10: More test coverage for functional tests
Bug fix:
PR9: Fixed some streamlit issues
PR11: Fixed predict scripts errors with multilabel tasks
PR12: Fixed some functional tests
