This stage of the project involves the evaluation of the model against the validation and test datasets.
Key Outcomes:
- Versions of model improvements and iterations captured.
Note, during the extraction post-processing will need to take place alongside application of business rules to ensure that the data is transformed into a state or format required for further downstream tasks and integration. The code accelerators below provide the scaffolding utilising simple python scripts to:
- Retrieve the storage containers that contain the test datasets
- Get the associated model to the issuer of the form via a lookup
- Sample a configurable number of test forms randomly
- OCR the test forms
- Evaluate the forms
Note, this assumes that a classification step has taken place before to infer the correlation between the issuer of the form and the trained model associated with that form type/layout. Have a look at the code accelerator Attribute Search for a simple approach that can help implement this.
Have a look at the code accelerator for evaluating a model with the Supervised Form Recognizer
Have a look at the code accelerator for evaluating a model with the Unsupervised Form Recognizer
Now refer to the Evaluation section to for the final stage of forms extraction.
Back to the Training section