DocumentAI - Tests vs training

Google Cloud Build with AI AI Solutions
1

Hello,

I don’t really understand the difference between documents used for tests or for training.

What is the ideal order of things?

  1. I create my custom processor.
  2. I create my labels.
  3. I insert 10 training documents.
  4. Then 10 test documents.
  5. And in the “evaluate and test” tab, I start a complete evaluation?

And if I find that it’s not precise enough, do I need to submit training or test documents?

Thank you for your help.

5 Likes
2

Hello @aksld ,

If you’re using a custom extractor, you can follow this guide to help you on the step-by-step instructions. Otherwise, if you need more information and a step-by-step guide on other Document AI processors, you may follow this link.

As for your question,

“And if I find that it’s not precise enough, do I need to submit training or test documents?”

Based on the documentation,

“If your documents all have a similar format (for example, a fixed form with very low variation), then fewer documents are required to achieve accuracy. The higher the variation, the more documents are required.”

Additionally, you could up train new processor versions to improve accuracy on your data, extract additional custom fields from your documents, and add support for new languages. However, not all specialized processors support up training.

I hope this helps.