Implementing Custom Classification Model in Document Intelligence

This codebase provides an example implementation of a custom classification model in 4 steps:

• Document Preparation (analyze_layout.py)
• Document Upload (upload_documents.py)
• Build Classifier (build_classifier.py)
• Classify Documents (classify_document.py)

Prerequisites

In order to complete this workshop, you will need to the following:

• Python 3.11 or higher (recommended using an Anaconda environment)
• Visual Studio Code
  • Python and Jupyter extensions
• Access to Azure Cognitive Services
• Access to an Azure Storage Container

Configuration

Before running the scripts, you need to set up your environment variables. Rename the .env.txt to .env file and include the following variables:

• AZURE_DOCUMENT_INTELLIGENCE_ENDPOINT: The endpoint to your Document Intelligence resource.
• AZURE_DOCUMENT_INTELLIGENCE_KEY: Your Document Intelligence API key.
• AZURE_STORAGE_CONNECTION_STRING: The connection string to your Azure Storage service
• AZURE_STORAGE_CONTAINER_NAME: The name of your Azure Blob Storage container
• TRAINING_DOCUMENTS: The path to your training documents
• TESTING_DOCUMENTS: The path to your testing documents
• CLASSIFIER_ID: The model ID of your Document Intelligence (wait until after running build_classifier.py)
• BASE_CLASSIFIER_ID: The model ID of your base classifier (edit only if you want to perform incremental training on an existing classifier)

Please replace the placeholders with your actual values.

Your TRAINING_DOCUMENTS folder should be structured as shown below:

📂TRAINING_DOCUMENTS
 ┣ 📂DocumentType1
 ┃ ┣ 📜trainingFile1.ext
 ┃ ┣ 📜trainingFile2.ext
 ┃ ┣ 📜trainingFile3.ext
 ┃ ┣ 📜trainingFile4.ext
 ┃ ┣ 📜trainingFile5.ext
 ┃ ┗ 📜...
 ┣ 📂DocumentType2
 ┣ 📂...

You must include AT LEAST 5 training files for each type of document you wish to train the model on.

Install packages

Install the required modules

pip install -r requirements.txt

analyze_layout.py

This script uses the Document Intelligence layout model to analyze your training files and create corresponding .ocr.json files. These files are saved locally alongside your training data files and will be uploaded when running the upload_documents.py script.

python analyze_layout.py

upload_documents.py

This script uploads labeled data to your Azure Blob Storage container.

python upload_documents.py

build_classifier.py

This scripts demonstrates how to build a classifier model.

python build_classifier.py

Remember to copy and paste the Classifier ID in .env

classify_document.py

This scripts demonstrates how to classify a folder of documents using a trained document classifier via

python classify_document.py

Resources

• Azure Document Intelligence service: https://learn.microsoft.com/en-us/python/api/overview/azure/ai-documentintelligence-readme?view=azure-python-preview#examples

• Documentation: https://github.com/Azure/azure-sdk-for-python/tree/main/sdk/documentintelligence/azure-ai-documentintelligence/samples