This topic provides a user guide for secure environment modeling. It explains how to create a secure environment project, log on to the secure environment, and customize a model.
1. Create a modeling project
Log on to the Fraud Detection console.
In the navigation pane on the left, choose to go to the Secure Environment Modeling page.
In the upper-left corner of the page, click Create Modeling Project. In the dialog box that appears, enter a Project Name and click OK to create the project.
2. Associate backtrack tasks
Custom modeling relies on modeling features. This step associates submitted feature backtrack tasks with the project. After the feature backtrack is complete, the features are automatically uploaded to the secure environment.
Prerequisite: Submit a backtrack task for modeling features. For more information, see Test in the product console.
In the Associate Backtrack Task section, click the Associate New Task button.
In the Associate Task dialog box, search for the feature backtrack task by its name and click OK.
3. Create a logon account for the secure environment
After you associate a backtrack task, you can view the machine preparation progress in the Secure Environment Status section.
Follow the on-screen instructions to enter a Phone Number for logon Captcha validation.
When the account is created and the backtrack features are ready, the Secure Environment Status section displays the logon information, including the account and initial password.
Modelers can then follow the "Operation guide" to log on to the secure environment and complete the modeling.
Secure environment operation guide
4. Export model files
After modeling, package the required model reports, scoring files (which must contain only the primary key and score), and model Predictive Model Markup Language (PMML) files. Place the package in the Desktop/download folder.
The packaged files required for deployment are as follows:
Model PMML files: Versions 4.3 and 4.4 are supported.
The list of features used for modeling.
Validation file: An offline model scoring file with a maximum of 5,000 records. The file must include the following fields: encrypted mobile number, lookback time, modeling features, and the offline model score. The offline model score field must be named prediction_score. If feature processing is involved, provide the processed results.
On the project product page, click the Synchronize Model Files button to initiate an export request.
After a system security review, which typically takes 1 to 2 business days, the files will appear in the Modeling Progress module.
5. Complete or close a modeling project
Before the modeling is complete, you can click the Close Project button in the upper-right corner of the details page. After you confirm the action in the dialog box, the secure environment is released.
After you close the project, the secure environment is automatically released. All files in the environment are destroyed and cannot be recovered.
This operation cannot be revoked. Proceed with caution.
After the modeling is complete, you can click the Complete Project button in the upper-right corner of the details page. After you confirm the action in the dialog box, the secure environment is released.
After you complete the project, the secure environment is automatically released. All files in the environment are destroyed and cannot be recovered.
This operation cannot be revoked. Proceed with caution.
6. Notes
Only one account can be logged in to the secure environment at a time.
Implement the feature processing logic in the DataFrameMapper of the pipeline.
If you encounter connection issues with the secure environment, contact your Alibaba Cloud representative for assistance.