AI-102 Question 411
Single answerYou are building a custom document intelligence model using Azure AI Form Recognizer. The model needs to extract invoice data (e.g., invoice numbers, dates, and total amounts) from a variety of invoice layouts. After training the model with labeled data, you want to validate its performance before publishing it for production use. Which step should you perform to ensure the model's accuracy before publishing?
- A
Use the 'Analyze' feature to test the model with unseen sample documents.
- B
Directly publish the model and monitor its performance in production.
- C
Train the model again with the same data to ensure consistency.
- D
Export the labeled data and manually review it for errors.
Show answer and explanation
Correct answer: A
Explanation
Testing the model with the 'Analyze' feature ensures it performs well on unseen documents and accurately extracts the required fields before it is deployed to production. This step is crucial in validating the model's effectiveness and avoiding potential errors in production scenarios.
- A. Correct.
Correct: The 'Analyze' feature allows you to test your trained model with new, unseen sample documents to validate its accuracy in extracting the required fields. This is a critical step before publishing.
- B. Incorrect.
Incorrect: Publishing the model without validation can lead to errors in production as you would not have verified its accuracy beforehand.
- C. Incorrect.
Incorrect: Retraining the model with the same data does not improve its accuracy or ensure it works on unseen data. Testing is required for validation.
- D. Incorrect.
Incorrect: Reviewing labeled data manually does not test the model's performance and only ensures data quality, not model accuracy.