AI-102 Question 413
Select 3You are a data scientist tasked with building a custom document processing solution for your company using Azure Form Recognizer. The goal is to extract invoice details such as invoice number, date, and total amount. After creating a custom model, you need to ensure that the model is well-trained, accurately tested, and ready for deployment to production. Which steps must you take to achieve this goal?
- A
Label a dataset of sample invoices and use it to train the custom model.
- B
Test the trained model using the same dataset used for training to evaluate performance.
- C
Provide a separate dataset of unseen invoices to test the model's accuracy.
- D
Publish the model to an endpoint for integration with production applications.
- E
Use only prebuilt models in Azure Form Recognizer and skip training a custom model.
Show answer and explanation
Correct answers: A, C, D
Explanation
To successfully train, test, and publish a custom document intelligence model using Azure Form Recognizer, you must first label a dataset to train the model. After training, testing with a separate dataset ensures that the model performs well on unseen data. Once the model has been validated, publishing it to an endpoint allows integration into production workflows. Prebuilt models lack the customization needed for specialized field extraction and are not suitable for this use case.
- A. Correct.
Correct: Labeling a dataset is a mandatory step when training a custom model in Azure Form Recognizer. This allows the model to learn how to extract specific fields like invoice number, date, and total amount.
- B. Incorrect.
Incorrect: Testing the model using the same dataset used for training leads to overfitting and does not provide an accurate measure of the model's performance.
- C. Correct.
Correct: Using a separate, unseen dataset is essential for testing the model's generalization and accuracy on new data.
- D. Correct.
Correct: Publishing the model to an endpoint is required to integrate it with production systems and make it accessible for real-world applications.
- E. Incorrect.
Incorrect: While prebuilt models can be useful, they are not applicable when extracting custom fields that are unique to your specific use case. Training a custom model is necessary in this scenario.