Train a learning instance
Use the staging environment to train a learning instance and extract data from sample documents. Review each field to verify whether the correct value is extracted and where necessary, map the correct field.
When you upload sample documents while creating a learning instance, the Classifier analyzes and categorizes these documents into logical groups. The system creates a bot for each document group, which you must train to extract data from specific fields in the sample documents.
The following steps provide an overview of the training process:
- Step 1: Open the Designer
- From the My Learning Instances page, open a learning instance and click Edit bot or Create bot to start training the learning instance.
- For an introduction to the components, see Introduction to the Designer.
- Step 2: Map the document fields
- In the Fields section, click each form field to verify whether the correct value is extracted. If the value is incorrect or if the field was not found (indicated by a grayed-out check icon next to the field), you must remap the field.
- Step 3: Map the check boxes (if applicable)
- If you configured the learning instance and uploaded sample documents
containing check boxes or option buttons, see Extract data from check boxes or check box groupsNote: Check boxes and check box groups can be extracted as either a form or table field depending on whether the check box appears one time or on a recurring basis in the document.
- Step 4: Map the table fields
- Review each table field to ensure that the correct column name is identified in the Column Value field. If IQ Bot did not find the correct Column Name, remap the column name by using the same approach as you did for the form fields. Then, configure the table settings. Map a table
- If the document contains more than one table, define the additional tables:
- Add all the table fields to the Additional table/repeated fields section of a learning instance. Edit a learning instance.
- Map the additional table fields. Add multiple tables in Designer.
To establish a relationship between the fields of the different tables, see Define one or more linked fields in a child table.
- Step 5: Review the training results
- Click See extraction results to check the extracted data from the document group. Preview extracted data
- If not all of the fields are extracted, try resizing the mapping areas for those fields.
- Alternatively, train the bot by using a different document. Change the training document for a group
- Step 6: Save the bot
- Click Save and go to next group to train the bot for the next document group.
- A menu appears, with the option to Cancel,
Save, or Save and send to
Production.Note: We recommend that you test the bot (see next step) before sending it to production.
- You can view the status of all the bots in all your learning instances from the BOTS tab. Manage IQ bots
- Step 7: Test bot extraction capabilities
- Navigate to the BOTS tab and click Test bot.
- The bot processes the remaining documents in the document group based on the trained document. To see the bot performance, review the values in the Learning instance summary page, particularly the straight-though processing (STP) metric. Review the learning instance details
- If the STP number is low, you can either retrain the bot, such as by resizing the mapping areas or train the bot on a different document. Change the training document for a group
- Step 8: Send the bot to production
- When you are satisfied with the extraction accuracy of the bot, switch it to production mode. Then, switch the learning instance to production mode. Once a learning instance is in production mode, IQ Bot is ready to begin processing documents in real time: classifying the uploaded documents to the document groups with bots in production and either extracting data or sending the document for manual validation.
- From the Learning Instance Details page, click the
switch at the top right to move the learning instance into production
mode. Set learning instance to ProductionNote: If you want to retrain a bot, both the learning instance and the bot must be in staging mode.