tarandm_analytics 1.2.1.dev2

Package links analytics in Python with TaranDM software.

TaranDM analytics

TaranDM analytics is a package with supportive functions for TaranDM decision manager software. Two main areas are covered:

• Preparation of predictive model deployment
• Preparation of dataset for predictive model development - attribute evaluator

An example notebook how to use tarandm_analytics is included in the package in tarandm_analytics/examples/tarandm_model_development.ipynb

Predictive model deployment

Strategies in TaranDM can contain predictive models. For compatibility, TaranDM requires specific format in which the model is deployed. tarandm_analytics package provides functions to make the deployment easy.

In TaranDM, predictive models are stored alongside with additional metadata. Those can be used for instance to monitor the model stability. Information about development sample and model performance is also stored amongst others.

After training the predictive model, steps to prepare the model for deployment would typically be:

1. Initialize ExportPredictiveModel from tarandm_analytics package.
2. Prepare monitoring data using get_monitoring_data method. This will calculate data to monitor stability through population stability index (PSI).
3. Prepare predictive model data for export to disk using prepare_predictive_model_data method.
4. Export model to disk using build_predictive_model method. Model is exported in zip format, that can be uploaded to TaranDM strategy in GUI.

Attribute evaluator

Attribute evaluator provides functions to create a dataset for predictive model development. Past requests are fetched from database and attributes defined in TaranDM attribute classes are evaluated. It uses the same code to evaluate attributes as the production environment, which eliminates potential mismatch in attribute definition during implementation to production.

User can either define past requests to be included in the dataset by listing decision ids directly or by defining business case and time range. Attribute classes to be evaluated are also defined by user.

To prepare the dataset:

1. Initialize EvaluateAttributes class.
2. List available attribute classes using get_attribute_classes.
3. List available business cases using get_business_cases.
4. Run evaluate method. Business case, time range or list of decision ids are provided as parameters of the method. Note that user is required to provide Git repository with strategies as well as credentials for the repository.
5. Once the attributes are evaluated, fetch the data using fetch_data_from_db method. It requires process ID as a parameter. This can be found in last_attribute_extractor_id property of EvaluateAttributes object.