Release Notes – Indico IPA v3.4
June 12, 2020 / Release Notes
Thank you for being a valued Indico user! We’re constantly making updates to our app and APIs, working on new features, and garnering feedback to be best in class for intelligent process automation. Have ideas on how to make our product even better – please let us know here!
- On Document Labeling
- A few quality of life improvements have been made to help with happy labeling.
- Some examples of changes include: a hotkey to move to the next example, truncation of long file names, keyboard shortcut to toggle between text highlighting and bounding box.
- Review is now Explain
- In anticipation of our early Q3 release of On Document Review (the ability to correct text extractions through our On Document Labeling interface), we are rebranding the current Review as Explain – the same functionality will be available just under a new name!
- Metrics for Sequence Models via API & in App
- For sequence models, we will now have token level metrics for precision, recall by label.
- Per sequence model, an overall F-score and the counts of the false positives, false negatives, and true positives by class.
- In Explain (formerly Review), these metrics will be available for all sequence models
- User Management Redesign
- Users can now be provisioned to the dataset on the Datasets page with their role, including any Indico registered user to be a Manager, Analyst or Labeler.
- In Teach, the users are now in a drop down to easily permission a “Labeler” to a specific task.
- Roles can now be changed without removing and re-adding the same user, and all the associated labels from a user will be maintained even if a user is removed from a task or if the role is changed.
- Workflows allow multiple docbots, processors, and models to be linked together in sequence to programmatically generate outputs by adding data to the workflow for submission – simply put: you put unstructured data in, we give you structured data out!
- We currently support the following workflow: PDF input that passes through OCR to a single classification model feeding to an annotation model. This is by far, the most common workflow we have seen to date.
- If there are more complex workflows, experimentation is encouraged, but in the words of our Frontend Engineer, “there may be dragons!” so please proceed cautiously and know you may run into features that are still under development.
- Once the workflow is created, you can submit new samples for processing via Python Client Library calls.
- Try It Out for Workflows
- A sample document can be run through a workflow to receive an output CSV containing all the predicted values and confidence scores for each.
- Try It Out supports the Workflow use cases that are supported via API listed above
- If you are using Try It Out for more complex workflows, some features may still be under development, but experimentation is encouraged!
- Try out a workflow for troubleshooting or validation easily from within the UI – just look in your Dataset, and this is available in the Workflow tab.
- General Improvements
- We’re always looking to improve the functionality of our app and our APIs.
- Some fixes we’ve made this release:
- Progress bar is back and better than ever!
- PDF Extraction will again provide table cell metadata on export.
- Better text extraction on document with low DPI (dots per inch).
- Updating a model name in Explain will now successfully save.
- Continuing updates to the API documentation at indico.io/docs
- Updates Java Client Libraries
- At Indico, we typically are a Python first organization, with quick follow on for both C# and then Java barring specific customer needs
Early Summer 2020
- Repair and validation workflow for annotation tasks
- Labels and predictions will have a review queue to validate and correct any text or annotation labeling errors prior to export from within the Indico app
- On Document Labeling for classification task, including ability to split filter datasets at the document, page, or paragraph for labeling
- Repair and validation for all models within a workflow
- SSO/Active Directory Integration
- Watch Folders
- Automate dataset upload through our watch folders by processing documents as they become available