Reverse modeling: Create models from physical tables

Modeling perspective

Dimensional modeling organizes model tables into three levels: Common layer, Application layer, and Uncategorized. The Common layer is used to build reusable unified metrics, dimensions, and detailed fact data, and supports management from either data domain or business category perspectives. The Application layer addresses business-specific statistical needs and supports only the business category perspective. After selecting a level, you can create and manage model tables in the corresponding directory tree.

Limitations

Reverse modeling is supported only for tables in the production environment of MaxCompute and E-MapReduce (EMR) Hive.

Modeling process

Procedure

1. Go to the Reverse Modeling page.

   1. Log on to the DataWorks console. In the target region, click Data Development and O&M > Data Modeling in the left-side navigation pane. Select a workspace from the drop-down list and click Go to Data Modeling.

   2. In the top navigation bar of the Data Modeling page, click Dimensional Modeling to go to the Dimensional Modeling page.

   3. In the left-side navigation pane of the Dimensional Modeling page, click Reverse Modeling to go to the Reverse Modeling page.

2. Start modeling.

   • If this is your first time using reverse modeling, click Start Now in the center of the Reverse Modeling page to start.

   • If you have used reverse modeling before, click Start Reverse Modeling in the upper-right corner of the Modeling Tasks page.

3. Configure a reverse modeling policy.

   Note

   Reverse modeling is irreversible. Once used to generate models, a policy cannot be modified. Plan your policy carefully based on your business requirements.

   1. Configure the policy settings.

      Parameter	Description
      Workspace	Select the DataWorks workspace that contains the tables you want to reverse model. Note You can select only DataWorks workspaces where the current user is a member. To select a different workspace, add the user as a member to that workspace. For more information, see Manage permissions on workspace-level services.
      Compute Engine Type	Reverse modeling is supported only for physical tables from MaxCompute and E-MapReduce (EMR) Hive projects in the production environment.
      Compute Engine Instance	Select the compute engine instance that contains the tables you want to reverse model.
      Table Name Matching Rule	Select a method to match tables in the target compute engine. The matched tables are then reverse modeled. The available methods are: Fuzzy Match: Enter a keyword to match all tables whose names contain that keyword. Exact Match: Enter the full, exact name of one or more tables. Note To specify multiple table names, separate them with a semicolon (;) without a space after the semicolon. If no tables are matched, the task fails and no models are generated.
      Data Layer of Model After Reverse Modeling	Common Layer: Select this layer to create fact tables, dimension tables, or aggregate tables. Application Layer: Select this layer to create application tables or dimension tables.
      Table Naming Rule	This feature is used to parse the names of tables that are matched by the Table Name Matching Rule. This process standardizes model names after reverse modeling and mounts the models to the corresponding data warehouse layers. The details are as follows: Parsing rule The system parses the names of matched tables and the underscores (_) within them. A table name can contain up to nine underscores. The segments between underscores can be configured to represent information such as Business Process, Data Domain, or Custom content. If a table name contains an identifier for a data warehouse layer, the generated model is automatically assigned to that layer. Note If a table name does not contain the required layer identifiers, such as Data Domain or Business Process, the model's layer assignment will be empty. You can assign the layer later in the Confirm Model Information step. Parsing method Table Name Checker: Select a pre-configured checker to parse table names. To learn how to create a checker, see Configure data warehouse layer checkers and Use checkers. Custom Rule: Create a custom naming convention by combining elements such as Business Process, Data Domain, Business Category, and Custom content to parse table names.
      Execution Method	Select a method for creating models: Full Update: DataWorks creates models for all matched tables in Dimensional Modeling. Select this method to regenerate all matched tables as new models. Note If a model for a matched table already exists in Dimensional Modeling, selecting Full Update deletes the existing model and creates a new one. Incremental Update: DataWorks performs the following actions on the matched tables: Identifies and filters out tables that already have corresponding models in Dimensional Modeling. Creates new models only for matched tables that do not already have a model in Dimensional Modeling. Select this method to avoid recreating existing, unchanged models.

   2. Click Create Model to parse the tables based on the policy.

4. Confirm model information.

   1. Confirm the model information.

      DataWorks generates initial models based on your policy. You can adjust information such as the Table Type and its associated Data Layer and Data Domain. You can also delete tables that you do not want to model.

   2. Click Generate Model to create the models.

5. View the results.

   After the models are generated, you can view the number of successfully created models of each type and the details of any failed models. Click Error Logs to quickly identify and troubleshoot issues.

   Note

   • Successfully generated models are automatically materialized to the corresponding compute engine. You do not need to publish them.

   • Successfully generated models are stored in Dimensional Modeling. You can go to the Dimensional Modeling page to view and manage them. For more information, see Model management.

View modeling tasks