v1.10.2

Improvements

• Introduced generate_model_card into write_json_files.py to allow for python models to work with planned model card tab in SAS Model Manager.

Bugfixes

• Allow for score code to impute NaN values in tables that have been loaded into SAS Model Manager.
• Fix issue where target_value was not being properly set during score code generation
• Updated pzmm_generate_requrirements_json.ipynb so the requirements file is generated properly.
• Added missing statistics to dmcas_fitstat.json file.

v1.10.1

Improvements

• Introduced ability to specify the target index of a binary model when creating score code.
  • index can be specified in pzmm.import_model.ImportModel.import_model()
  • Relevant examples updated to include target_index.

Bugfixes

• Reworked write_score_code.py to allow for proper execution of single line scoring.
• Added template files for assess_model_bias.py to allow for proper execution

v1.10.0

Prep for release

v1.9.4

Improvements

• Created pytest fixture to begin running Jupyter notebooks within the GitHub automated test actions.
• Updated examples:
  • Custom KPI and model parameters example now checks for the performance job's status.
  • Update H2O example to show model being published and scored using the "maslocal" destination.
  • Updated models to be more realistic for pzmm_binary_classification_model_import.ipynb.

Bugfixes

• Adjust pzmm.ScoreCode.write_score_code() function to be compatible with future versions of pandas.
• Reworked H2O section of pzmm.ScoreCode.write_score_code() to properly call H2OFrame values.
• Fixed call to pzmm.JSONFiles.calculate_model_statistics() in pzmm_binary_classification_model_import.ipynb.

v1.9.3

Improvements

• Refactored gitIntegration.py to git_integration.py and added unit tests for better test coverage.

Bugfixes

• Fixed issue with ROC and Lift charts not properly being written to disk.
• Fixed JSON conversion for Lift charts that caused TRAIN and TEST charts to be incorrect.
• Fixed issue with H2O score code and number of curly brackets.
• Updated score code logic for H2O to account for incompatibility with Path objects.
• Fixed issue where inputVar.json could supply invalid values to SAS Model Manager upon model import.
• Fixed issue with services.model_publish.list_models, which was using an older API format that is not valid in SAS Viya 3.5 or SAS Viya 4.

v1.9.2

Improvements

• Add recursive folder creation and an example.
• Add example for migrating models from SAS Viya 3.5 to SAS Viya 4.

Bugfixes

• Fixed improper json encoding for pzmm_h2o_model_import.ipynb example.
• Set urllib3 < 2.0.0 to allow requests to update their dependencies.
• Set pandas >= 0.24.0 to include df.to_list alias for df.tolist.
• Fix minor errors in h2o score code generation

v1.9.1

Improvements

• Updated handling of H2O models in sasctl.pzmm.
  • Models are now saved with the appropriate h2o functions within the sasctl.pzmm.PickleModel.pickle_trained_model function.
  • Example notebooks have been updated to reflect this change.

Bugfixes

• Added check for sasctl.pzmm.JSONFiles.calculate_model_statsistics function to replace float NaN values invalid for JSON files.
• Fixed issue where the sasctl.pzmm.JSONFiles.write_model_properties function was replacing the user-defined model_function argument.
• Added NpEncoder class to check for numpy values in JSON files. Numpy-types cannot be used in SAS Viya.

v1.9.0

Improvements

• sasctl.pzmm refactored to follow PEP8 standards, include type hinting, and major expansion of code coverage.
  • sasctl.pzmm functions that can generate files can now run in-memory instead of writing to disk.
• Added custom KPI handling via pzmm.model_parameters, allowing users to interact with the KPI table generated by model performance via API.
  • Added a method for scikit-learn models to generate hyperparameters as custom KPIs.
• Reworked the pzmm.write_score_code() logic to appropriately write score code for binary classification, multi-class classification, and regression models.
• Updated all examples based on sasctl.pzmm usage and model assets.
  • Examples from older versions moved to examples/ARCHIVE/vX.X.
• DataStep or ASTORE models can include additional files when running tasks.register_model().

Bugfixes

• Fixed an issue where invalid HTTP responses could cause an error when using Session.version_info().

v1.8.2

Improvements

• folders.get_folder() can now handle folder paths and delegates (e.g. @public).

Bugfixes

• Fixed an issue with model_management.execute_model_workflow_definition() where input values for
  workflow prompts were not correctly submitted. Note that the input= parameter was renamed to
  prompts= to avoid conflicting with the built-in input().

v1.8.1

Changes

• Adjusted workflow for code coverage reporting. Prepped to add components in next release.
• Added generate_requirements_json.ipynb example.

Bugfixes

• Fixed improper math.fabs use in sasctl.pzmm.writeJSONFiles.calculateFitStat().
• Fixed incorrect ast node walk for module collection in sasctl.pzmm.writeJSONFiles.create_requirements_json().