Workflow

This following figure depicts the workflow in the Model Fabric application to build, train, register and deploy the models.

 Model Fabric Workflow

This section describes the workflow in the Model Fabric application for the following types of models:

  • BuiltIn Models: Models that are built and trained inside the Model Fabric application.

  • Imported Models: Models that are built and trained outside the Model Fabric application and then uploaded into the Model Fabric application.

BuiltIn model - Workflow

  1. The first step to build a model in Model Fabric application, starts at Data Exploration module, where you upload data from CDL, ICM or local file (CSV/XLS).

  2. Analyze how the data is distributed, select data variables and instances, get a statistical summary and visualize the dataset in various basic and custom plots.

  3. Identify missing data and outliers.

  4. Once you have explored and understood the data, go to the Data Pre-Processing and Model Recommendation module, where you create experiment for the dataset and select target variables.

    There are two modes available in the Model Fabric application to create the experiment and build the model accordingly: Express Mode (used by functional users) and Advanced Mode (used by data scientists and citizen data scientists).

    In the Express Mode, the Model Fabric solution creates the experiment of the selected dataset and processes the data based on the Model Recommender settings defined in the App Settings module. You can view the pre-processed data in the various tabs of the Data Pre-Processing page.

    However, data scientists and citizen data scientists can use the Advanced Mode to create the experiment and edit the pre-processed data using their data science knowledge.

  5. Pre-process the data by applying Missing Treatment, Outlier Treatment, Feature Engineering and Scaling methods for each variable in the dataset.

  6. Define the Model Settings to train and build the model (by selecting Target Variable, Train-Test Split ratio, Split Type, Algorithms, K-Fold Validation, Hyper Parameter Tuning Timeout per model (in seconds) and Metrics).

  7. View the Model Metrics and Model Plots of the model built by the Model Fabric solution.

  8. Go to the Model Registry and Deployment module, where you register the trained model.

  9. Deploy the model.

  10. Test the model.

  11. Configure the deployed model in the Model Monitoring module, to enable model monitoring and define monitoring schedules to monitor the model.

Imported Models - Workflow

  1. The workflow of a model that is built and trained outside the Model Fabric application starts at the Model Registry and Deployment module, where you add and register the model (ONNX and Non-Onnx).

  2. Add/edit the configuration of the model.

  3. Deploy the model.

  4. Test the model.

 

Related Topics: