Overview
Once a new Field ID or Table ID model is trained or uploaded, before uploading the model, you can evaluate the projected automation based on the Field Identification Target Accuracy setting for Field ID models and Table Identification Target Accuracy setting for Table ID models. If there is already a live model, you can also compare the projected performance from the current live model to the trained candidate model to decide which one you would like to deploy.
To view model training results, navigate to Library > Models and click on the specific model that you would like to evaluate.
Automation vs. target accuracy
The Live Model charts on a Model Details page display how Field ID target accuracy and Table ID target accuracy affect automation in the Semi-structured layout models that are currently live.
Projections for Field ID models
Target accuracy – Your desired accuracy for the identification of fields. For more information about this value, see Flow Settings.
Automation – The percentage of documents where all fields are expected to be automatically identified.
Projections for Table ID models
Target accuracy – Your desired accuracy for the identification of table cells. For more information about this value, see Flow Settings.
Automation – The percentage of documents where all tables are expected to be automatically identified.
Cell automation – The percentage of table cells that are expected to be automatically identified.
Comparing training results
After you train a new Semi-structured layout model, you can compare the projected performance of the live model against that of the new model in the Field ID Model and Table ID Model charts. Note that projected model performance (e.g. accuracy and automation) can increase by adding additional QA documents.
To help compare the results of your models, the chart plots out automation levels at every accuracy point. Simply hover your cursor over any point on the chart to see more information.
Deploying the better model
If the projected metrics for the new model are better than the live model (e.g., similar Projected Accuracy with greater Projected Automation), deploy the Candidate model.
However, if the projected metrics for the new model are worse than the live model (e.g., Projected Accuracy and/or Projected Automation have significantly decreased), then simply keep the current model. To improve Field ID model performance, perform more QA and then run training on that new data. Furthermore, consider increasing your QA sampling rate.