CHANGELOG.md

v1.11.0 (2024-10-29)

Changes

• Added score_definition.py and score_execution.py to allow for score testing within SAS Model Manager
  • Included optional use of CAS Gateway for faster scoring. Only available in environments where Gateway scoring is properly set up.
• Added ability to include data pre-processing function within python score code using the preprocess_function argument.

Bugfixes

• Fixed issue where settings file was improperly imported in some score code files.

v1.10.7 (2024-10-02)

Changes

• Due to licensing restrictions, the sasctl package will no longer be available through Anaconda.

Bugfixes

• Fixed a bug that caused an error when performing SSL verification without a CA bundle specified.

v1.10.6 (2024-08-26)

Improvements

• Refactor tasks.py to utilize sasctl.pzmm functions.
• Add model_info class to better capture model information.

v1.10.5 (2024-08-01)

Buxfixes

• Updated write_json_files.py to allow for better support for prediction models
• Fixed issues relating to model card support.

v1.10.4 (2024-07-08)

Improvements

• Added example Jupyter notebook for OpenAI models.

Buxfixes

• Dropped support for Python 3.6 and Python 3.7, as those are no longer officially supported versions.
• Added dmcas_misc.json template file for model card generation.
• Updated generation of ModelProperties.json to allow for model card generation immediately upon upload.

v1.10.3 (2024-04-12)

Bugfixes

• Updated all examples to use current versions of sasctl functions
• Fixed bug in generate_model_card that threw an error when trying to generate the dmcas_misc.json file

v1.10.2 (2024-04-10)

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 (2023-08-24)

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 (2023-08-31)

Improvements

• write_score_code.py refactored to include ability to run batch scoring.
• Added handling for TensorFlow Keras models.
• Updated project creation to automatically set project properties based on contained models.
• Included capability to assess biases of a model using CAS FairAITools using pzmm.write_json_files.assess_model_bias().
• Added custom KPI support for H2O, statsmodels, TensorFlow, and xgboost.
• Updated examples:
  • Added example walking through the creation process of a simple TensorFlow Keras model.
  • Added example detailing the usage of pzmm.write_json_files.assess_model_bias() for a simple regression model
  • Updated pzmm_custom_kpi_model_parameters notebook to have correct parameter casing.

v1.9.4 (2023-06-15)

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 (2023-06-08)

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 (2023-05-17)

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 (2023-05-04)

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 (2023-04-04)

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 (2023-01-30)

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().
• Fixed an issue with pzmm.importModel.model_exists() where project versions were incorrectly compared, resulting in improper behavior when the project version already existed.
  • Better handling for invalid project versions included.

v1.8.1 (2023-01-19)

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().

v1.8.0 (2022-12-19)

Improvements

• Added Session.version_info() to check which version of Viya the session is connected to.
• Updated the properties= parameter of model_repository.create_model() to accept a dictionary containing custom property names and values, and to correctly indicate their type (numeric, string, date, datetime) when passing the values to Viya.
• Added services.saslogon for creating and removing OAuth clients.
• Added pzmm.JSONFiles.create_requirements_json() to create the requirements.json file for model deployment to containers based on the user's model assets and Python environment.

Changes

• Deprecated core.platform_version() in favor of Session.version_info().
• A RuntimeError is now raised if an obsolete service is called on a Viya 4 session (sentiment_analysis, text_categorization, and text_parsing)
• Replaced the JSON cassettes used for testing with compressed binary cassettes to save space.
• Updated the testing framework to allow regression testing of multiple Viya versions.
• Refactored the authentication functionality in Session to be more clear and less error prone. Relevant functions were also made private to reduce clutter in the class's public interface.
• Began refactor for sasctl.pzmm to adhere to PEP8 guidelines and have better code coverage.

Bugfixes

• Fixed an issue with register_model() that caused invalid SAS score code to be generated when registering an ASTORE model in Viya 3.5.
• Fixed a bug where calling a "get_item()" function and passing None would throw an error on most services instead of returning None.
• Fixed a bug that caused the authentication flow to be interrupted if Kerberos was missing.

v1.7.3 (2022-09-20)

Improvements

• Refactor astore model upload to fix 422 response from SAS Viya 4
• ASTORE model import now uses SAS Viya to generate ASTORE model assets
• Expanded usage for cas_management service (credit to @SilvestriStefano)

Bugfixes

• ASTORE model import no longer returns a 422 error
• Fix improper filter usage for model_repository service
• Fix error with loss of stream in add_model_content call for duplicate content
• Update integration test cassettes for SAS Viya 4

v1.7.2 (2022-06-16)

Improvements

• Added a new example notebook for git integration
• Added a model migration tool for migrating Python models from Viya 3.5 to Viya 4
• Improved handling of CAS authentication with tokens

Bugfixes

• Fixed git integration failure caused by detached head
• Fixed minor bugs in score code generation feature
• Fixed 500 error when importing models to Viya 4 with prewritten score code
• Fixed incorrect handling of optional packages in pzmm

v1.7.1 (2022-04-19)

Bugfixes

• Removed linux breaking import from new git integration feature
• Various minor bug fixes in the git integration feature

v1.7.0 (2022-04-07)

Improvements

• Added Git integration for better tracking of model history and versioning.
• Added MLFlow integration for simple models, allowing users to import simple MLFlow models, such as sci-kit learn, to SAS Model Manager

v1.6.4 (2022-04-07)

Bugfixes

• Fixed an issue where folders.create_folder() would attempt to use root folder as parent if desired parent folder wasn't found. Now correctly handles parent folders and raises an error if folder not found.

v1.6.3 (2021-09-23)

Bugfixes

• Fix an issue where pzmm.ZipModel.zipFiles() threw an error on Python 3.6.1 and earlier.

v1.6.2 (2021-09-09)

Bugfixes

• Fixed an issue with register_model() where random forest, gradient boosting, and SVM regression models with nominal inputs where incorrectly treated as classification models.

v1.6.1 (2021-09-01)

Improvements

• model_repository.add_model_content() will now overwrite existing files instead of failing.

Bugfixes

• PagedList.__repr__() no longer appears to be an empty list.

v1.6.0 (2021-06-29)

Improvements

• Session now supports authorization using OAuth2 tokens. Use the token= parameter in the constructor when an existing access token token is known. Alternatively, omitting the username= and password= parameters will now prompt the user for an auth code.

Changes

• current_session now stores & returns the most recently created session, not the first created session. This was done to alleviate quirks where an old, expired session is implicitly used instead of a newly-created session.
• Removed deprecated raw= parameter from sasctl.core.request().
• Dropped support for Python 2.

v1.5.9 (2021-06-09)

Bugfixes

• Fixed an issue that caused score code generation by pzmm module to fail with Viya 3.5.

v1.5.8 (2021-05-18)

Bugfixes

• SSL warnings no longer repeatedly raised when verify_ssl=False but CAS_CLIENT_SSL_CA_LIST is specified.
• model_repository.delete_model_contents() no longer fails when only one file is found.

Improvements

• All delete_*() service methods return None instead of empty string.
• All get_*() service methods issue a warning if multiple items are found when retrieving by name.

v1.5.7 (2021-05-04)

Bugfixes

• Fixed an import issue that could cause an error while using the pzmm submodule.

v1.5.6 (2021-04-30)

Improvements

• PagedList handles situations where the server over-estimates the number of items available for paging.
• The version of SAS Viya on the server can now be determined using sasctl.platform_version().

Bugfixes

• Reworked the model_repository.get_repository() to prevent HTTP 403 errors that could occur with some Viya environments.

v1.5.5 (2021-03-26)

Bugfixes*

• Fixed an issue with JSON parsing that caused the publish_model task to fail with Viya 4.0.

v1.5.4 (2020-10-29)

Improvements

• Added the as_swat method to the Session object, allowing connection to CAS through SWAT without an additional authentication step.

Changes

• Integrated PZMM into Session calls and removed redundant function calls in PZMM.
• ROC and Lift statistic JSON files created by PZMM are now generated through CAS actionset calls.
• Updated the PZMM example notebook, FleetMaintenance.ipynb, to include integration of PZMM with sasctl functions.

Bugfixes

• Reworked the model_repository.get_repository() to prevent HTTP 403 errors that could occur with some Viya environments.

v1.5.3 (2020-06-25)

Bugfixes

• Added PZMM fitstat JSON file to manifest.

v1.5.2 (2020-06-22)

Improvements

• PZMM module moved from a stand-alone repository to a sasctl submodule.
• Introduced deprecation warnings for Python 2 users.

v1.5.1 (2020-4-9)

Bugfixes

• Fixed PyMAS utilities to correctly work functions not bound to pickled objects.
• Model target variables should no longer appear as an input variable when registering ASTORE models.

v1.5 (2020-2-23)

Improvements

• Registered Python models will now include both predict and predict_proba methods.
• Added a new Relationships service for managing links between objects.
• Added a new Reports service for retrieving SAS Visual Analytics reports.
• Added a new Report_Images service for rendering content from reports.
• Additional metadata fields are set when registering an ASTORE model.
• Collections of items should now return an instance of PagedList for lazy loading of results.
• Module steps can now be called using module.step(df) where df is the row of a DataFrame or Numpy array.
• register_model sets additional project properties when registering an ASTORE model.

Changes

• Replaced the raw parameter of the request methods with a format parameter, allowing more control over the returned value.
• The get_file_content method of the Files service now returns the actual content instead of the file metadata.
• JSON output when using sasctl from the command line is now formatted correctly.

Bugfixes

• model_publish.delete_destination now works correctly.

v1.4.6 (2020-1-24)

Bugfixes

• Fixed an issue where the REQUESTS_CA_BUNDLE environment variable was taking precedence over the verify_ssl parameter.

v1.4.5 (2019-12-5)

Changes

• Saving of package information can now be disabled using the record_packages parameter of register_model.

Bugfixes

• Added support for uint data types to the register_model task.
• Fixed an issue where long package names caused register_model to fail.
• Session creation now works with older versions of urllib3.

v1.4.4 (2019-10-31)

Bugfixes

• Match performance definitions based on project instead of model.

v1.4.3 (2019-10-28)

Bugfixes

• Model versioning now works correctly for Python models
• Fixed an issue where None values in Python caused issues with MAS models.

v1.4.2 (2019-10-23)

Bugfixes

• Fixed project properties when registering a model from ASTORE.
• Fixed model metadata when registering a datastep model.

v1.4.1 (2019-10-17)

Bugfixes

• Fixed an issue where string inputs to Python models were incorrectly handled by DS2.

v1.4 (2019-10-15)

Changes

• PyMAS.score_code now supports a dest='Python' option to retrieve the generated Python wrapper code.
• register_model task includes a python_wrapper.py file when registering a Python model.
• Improved error message when user lacks required permissions to register a model.

Bugfixes

• Fixed an issue with CAS/EP score code that caused problems with model performance metrics.

v1.3 (2019-10-10)

Improvements

• Added update_performance task for easily uploading performance information for a model.
• New (experimental) pyml2sas sub-package provides utilities for generating SAS code from Python gradient boosting models.
• New (experimental) methods for managing workflows added to model_management service.

Changes

• register_model task automatically captures installed Python packages.
• All list_xxx methods return all matching items unless a limit parameter is specified.
• Improved API documentation.
• Updated full_lifecycle example with performance monitoring.

v1.2.5 (2019-10-10)

Changes

• Registering an ASTORE model now creates an empty ASTORE file in Model Manager to be consistent with Model Studio behavior.

Bugfixes

• microanalytic_score.define_steps now works with steps having no input parameters.
• Fixed an issue where score code generated from an ASTORE model lacked output variables.

v1.2.4 (2019-9-20)

Bugfixes

• model_repository.get_model_contents no longer raises an HTTP 406 error.

v1.2.3 (2019-8-23)

Changes

• put request will take an item parameter that's used to automatically populate headers for updates.

Bugfixes

• Convert NaN values to null (None) when calling microanalytic_score.execute_module_step.

v1.2.2 (2019-8-21)

Bugfixes

• register_model task should now correctly identify columns when registering a Sci-kit pipeline.

v1.2.1 (2019-8-20)

Improvements

• Added the ability for register_model to correctly handle CAS tables containing data step score code.

v1.2.0 (2019-8-16)

Improvements

• Added create_model_version and list_model_versions to model_repository
• Added an explicit ValueError when attempting to register an ASTORE that can't be downloaded.
• Added start and limit pagination parameters to all default list_* service methods.
• Added create_destination, create_cas_destination and create_mas_destination methods for model_publish service.

Changes

• Session.add_stderr_logger default logging level changed to DEBUG.

Bugfixes

• Fixed an issue where model_repository did not find models, projects, or repositories by name once pagination limits were reached.

v1.1.4 (2019-8-16)

Bugfixes

• The register_model task now generates dmcas_epscorecode.sas files for ASTORE models.

v1.1.3 (2019-8-14)

Bugfixes

• Fixed problem causing register_model task to include output variables in the input variables list.

v1.1.2 (2019-8-12)

Improvements

• CAS model table automatically reloaded on publish_model task.

Bugfixes

• Fixed DS2 score code for CAS that was generated when registering a Python model.
• PyMAS.score_code(dest='ESP') corrected to dest='EP'
• Fixed an issue where long user-defined properties prevented model registration.

v1.1.1 (2019-8-6)

Bugfixes

• Fixed an issue where usernames were not parsed correctly from .authinfo files, resulting in failed logins.

v1.1.0 (2019-8-5)

Improvements

• Added update_module and delete_module methods to MAS service.

Changed

• Added replace parameter to sasctl.tasks.publish_model
• Session hostname's can now be specified in HTTP format: 'http://example.com'.

Bugfixes

• Renamed microanalytic_store service to microanalytic_score

v1.0.1 (2019-07-31)

Changed

• Exceptions moved from sasctl.core to sasctl.exceptions
• SWATCASActionError raised if ASTORE cannot be saved during model registration.
• Improved handling of MAS calls made via define_steps()

v1.0.0 (2019-07-24)

Changed

• services are now classes instead of modules. Imports of services in the format import sasctl.services.model_management as mm must be changed to from sasctl.services import model_management as mm.
• host and user parameters of Session renamed to hostname and username to align with SWAT.
• Only InsecureRequestWarning is suppred instead of all HTTPWarning

Improvements

• Added copy_analytic_store method to model_repository service
• AuthenticationError returned instead of HTTPError if session authentication fails.

v0.9.7 (2019-07-18)

Improvements

• public_model task also defines methods mapped to MAS module steps when publishing to MAS.
• SSL verification can be disable with SSLREQCERT environment variable.
• CAs to use for validating SSL certificates can also be specified through the SSLCALISTLOC environment variable.
• Added execute_performance_task

Changes

• Updated method signature for create_performance_definition in Model Manager.

Bugfixes

• register_model task no longer adds rc and msg variables from MAS to the project variables.

v0.9.6 (2019-07-15)

Initial public release.