Edit

Version 10.2.6

Release date

September 5, 2022

Release 10.2.6 aims at enhancing your overall product user experience. The scope of these improvements includes:

• Internet Explorer mode support for Edge browsers allows you to run legacy and modern apps via a single browser at your convenience.
• Enhanced model management experience enables you to get your models up and running faster by leveraging the revamped navigation and extended dataset management functions in Control Tower's AutoML user interface.
• Improved role management helps create a truly multi-tenant environment and step up security with newly introduced roles and permissions in the user management system, Control Tower, and WorkSpace.

WorkFusion has also extended the novelty MT Designer component with the hotkey tagging feature, bringing you the same level of tagging convenience as with tasks designed in the old way.

The following Product upgrades will specifically benefit Engineers:

• Extended ML development capabilities with enhanced support for pure Python projects, improved definition generators, and stabilized model training performance
• Optimized BEP performance due to eliminated OCR Worker health checks
• Structured OCR output storage and extended OCR export formats
• RPA Server browser drivers will now be updated automatically
• Stabilized ODF performance due to issues resolved with unclosed Data Store connections
• Easier SSO access to MinIO

Traditionally, we have also updated specific third-party components to fix known vulnerabilities and eliminate the newly found bugs.

New roles and permissions

This release refines user roles and permissions to make user management within IA Cloud Enterprise even more flexible and granulated. The main idea is to segregate administrative rights to address two different tasks: set up roles within your organization and onboard new users. The segregation prevents administrators from granting themselves excessive authorization and protects your organization from possible harm.

For the User Management UI, the segregation entailed adding the following roles:

• wf-manage-groups-only. A grantee can create, rename, move, and delete groups with this pre-defined role. However, they cannot manage users and modify or delete any groups they belong to. To modify their group, the grantee also needs the manage-users or manage-realm role.
• wf-manage-realm-roles-only. This pre-defined role allows the grantee to create, modify, and delete realm roles. Unlike manage-realm, the role will not let the grantee modify or delete their role.
• wf-manage-client-roles-only. The role lets the grantee create, modify, and delete client roles. However, it does not allow any operation with the client directly or let them modify or delete client roles. To modify their own client roles, grantee still need manage-clients or manage-realm roles.

Control Tower offers the new Restricted Manage Control Tower Permissions privilege, allowing grantees to manage CT permissions for all roles, except for the ones they may have already. Also, Manage Users and Groups was renamed to Manage Control Tower Permissions.

WorkSpace features the new Manage other user roles permission, allowing grantees to manage WorkSpace permissions for all roles, except for the ones they may have. Also, Manage roles was renamed to Manage all roles.

For details, refer to Segregate Administrator permissions.

Internet Explorer mode for backward compatibility

Do you still have the good old Internet Explorer in use? Racking your brain, wondering how to make it compatible with advanced modern applications? WorkFusion has a solution: we enhanced our Product with the Internet Explorer mode support in Edge browsers implemented by Microsoft.

The feature allows managing legacy and modern web applications via a single browser. This means fewer cumbersome workarounds for your developers while coding Bot Tasks or working with RPA Recorder. For your business, it is a chance to save some more expenses on resolving backward compatibility issues. For usage details, refer to Support Edge in IE mode.

The Open Website Action in the Desktop RPA now contains two additional Browser options: Edge and Edge (IE Compatibility Mode).

Enhanced model management user experience

Non-engineer users can enjoy the benefits of a more intuitive model management flow and extended dataset handling capabilities in the Control Tower's AutoML interface:

• Move datasets across environments or reuse them for faster model training with new export, import, and copy functionality.
• Save time to view and modify model parameters and see how changes work with improved navigation. Straight from the model details page, you can access testing and modifying flows or go to associated training runs and test sets.
• Get better control over your model test and training data with export of extended "data point" and "per document" statistics via the test or model details pages. However, the extended per document statistics are available only for models developed with ML SDK version 10.2.6 and higher.

To explore new features in detail, refer to AutoML UI documentation.

Improvements

Manual Task Designer

• For novelty Manual Task Designer, we have enabled hotkey tagging: you can set up hotkeys to tag specific data points while designing a task.

This new feature not only levels up user experience with document tagging but saves you money and human resources: manual handling time is lower, and productivity is higher. For details, refer to Design Manual Tasks via Task Designer Operation.

AutoML

• Previously, definition generators searched for normalizers and post processors only in the workfusion.com package. With this release, the search scope has been extended to cover custom components outside the package, allowing you to apply customized model configurations.

• To improve Python integration and extend the possibilities of working with pure Python projects, we have introduced the following enhancements:

• In projects, developers can configure the resources directory to copy to the output model. The model.yml configuration file, build.py, and HypermodelArtifactBuilderwere extended accordingly.

• Earlier, during local model runs, only content used to be parsed for documents. Now, the runner provides parsing of attributes and gold values for JSON and CSV documents.

• The document parser was improved to accept all input types, whereas, previously, models in textParts expected the input to be either text, HTML, or XML.

• Pre-checks for Python environment installation now include validation for the env.zip file names. The validation is based on the python*.zip pattern.

• Wrapper archetypes containing pure ML SDK models are now ODF-compatible bundles. Developers can use them the same way as standard archetypes in ODF-based projects.

• For wrapped Python-based models, we offer refined log processing. Any errors occurring at Python start are printed as INFO.

• For pure Python archetypes, our engineers improved the handling of build script errors. Build.py stops should any of its subprocesses (pip install, maven install, and so on) fail.

• The default minDocs value for training Information Extraction and Classification models was updated to ensure that minimum test set requirements were met. If the minimum limit is not satisfied, an exception is generated. The solution helps you to avoid feature issues, thus stabilizing the performance of model training.

RPA

• The browser drivers on the RPA Server are now updated automatically without any interruptions to the environment operation, provided your Internet connection is stable. The new feature is available for Chrome and Microsoft browsers only. For instructions on enabling the feature, refer to Automatic updates of browser drivers.

OCR

• The OCR Worker health checks triggered by OCR REST were removed since all BEP workers already have health checks implemented via the Spring Boot actuator. Related OCR health check parameters are disabled by default. As a result, the BEP resources consumed by OCR Worker health checks were freed for other tasks on production, thus boosting the overall BEP performance.
• Now, OCR Bridge and OCR REST support the ALTO export format, thus expanding your options for handling the OCR output.
• In previous versions, you could not configure a precise location for your OCR output in MinIO. For more efficient data access and management, a parameter was added to the OCR Bridge configuration to set up specific folders in MinIO to store specific OCRed files. For details, refer to Add OCR Bridge step.

ODF 2

• For ODF2, a mechanism was implemented to check automatically for any unclosed Bot Task connections to Data Stores. ODF2 fails further execution with an explicit diagnostic message if some connections are left open after task execution. The improvement resolves the problem of bots failing to execute due to exceeded maximum limit of the Data Store connection pool.
• Your developers can now leverage the capability of running multiple Digital Worker variations with SPOKE tests. The Business Process testing framework now allows you to get variation_id after a newly created variation bundle is deployed and leverage it to prepare Data Store states before testing. For details, refer to Business Process integration testing.

Bug fixes

AutoML

• Fixed the AutoML Model Service issue releated to the data set size limitation of 512 items. This bug could prevent AutoML model training from starting.
• Resolved the issue when model packages generated from common archetypes missed meta-info.json files. Now, the model is generated successfully with the meta-info.json file available in the resources folder of the package module.
• Fixed the issue with multiple training requests started successfully with the same ID. Now, only the first request is started, whereas all further responses return a validation error stating that the model with training ID already exists.
• Fixed the issue related to the log level passed to Maven during pure Python build.
• Resolved the problem with label removal on the Labels tab of the dataset details page in the AutoML UI. Now, when a user clicks the Remove button and opens another page, all labels are removed.
• Fixed the broken outline of the Line item with a child label on AutoML > Dataset > Label.
• Resolved the AutoML UI issue when a new model training set up after creating a previous one and closing the training wizard contained an incorrect label configuration.
• Fixed the labeling flow issue in the AutoML UI. Now, the labeling flow is enabled for document datasets even if all labels on the Labels tab are deleted.
• Fixed the issue with the upload bar on the Document tab of the dataset details page. Now, if the upload fails, you get a message, and the upload bar disappears.
• Fixed the issue with the incorrect display of the document labeling status. Now, if you select a document for labeling but place no values on the configuration page, the document is identified and marked as Unlabeled in the document list.
• Eliminated the problem with incorrectly applied normalizer configurations in the AutoML UI. The fix lets models extract dates correctly in the required MMM-DD-YYYY format.
• Fixed the issue with unnecessary debug exceptions shown during pure Python builds. Now, as you build custom Python models, any debug messages coming directly from the Reflections library are suppressed, whereas the log level is passed explicitly to Maven.

Control Tower

• Fixed the issue when it was impossible to delete the USE_CASE_VERSION field from the meta-info.json file of a Bot Config Bundle project. Thanks to the fix, ODF 2 installation artifacts, and auxiliary AutoML Business Processes no longer appear in the Digital Worker list.
• Fixed the issue with a lost class loader after custom scraper preparation. When running tasks with ICL BCB, custom scraper preparations were required and demanded the replacement of the class loader with a plugin. For some reason, it became lost occasionally, but now the original thread class loader is restored after a task is finished.

ODF 2

• Fixed the failure that appeared during the compilation of an ODF 2 Bot Task containing space in the filename.
• Fixed the issue with Keycloak blocking a user on behalf of which SPOKE authenticated with bad credentials.
• Eliminated SPOKE issues with handling Business Process bundles created using the Enterprise Edition v10.2.5 and containing a manual step. Now, bundles are uploaded without issues.
• Fixed the failure of a Business Process containing InputMonitorTask due to the missing variation_id column.

WorkSpace

• Fixed the issue when drag-and-dropping columns was not working correctly on the Assignment list page.

Updates

• Azul JDK used in the installer was updated to version 8u342.
• ELK stack was upgraded to v7.17.4 to fix vulnerability issues in Kibana.
• OpenSSL libraries were upgraded to v1.0.2k-25.