Azure Custom Classification Model - How set "split mode" with Python?

Question

Azure Custom Classification Model - How set "split mode" with Python?

Roberto Araujo Filho 115

I have created a Custom Classification Model using Azure Document Intelligence Studio that works fine. But it classifies page by page of a document and I would like to get just one classification for the whole document.

The Document Intelligence Studio´s interface provides a buttom (Analyse options) where I can set this condition when classifying using this interface (as you can see below).

User's image

But I need to set this option to "none" inside my python code. I´ve tried some solutions like this:

poller = document_analysis_client.begin_classify_document(classifier_id, document=f, split="none")

Unfortunately the 'begin classify document' method doesn't accept this argument (split) and I couldn't find a way to configure it by looking in the Azure SDK for Python documentation.

I'll be happy if anyone can help.

Konstantinos Passadis 19,586 Reputation points MVP

2024-01-05T23:32:18.69+00:00
Hello @Roberto Araujo Filho !

Welcome to Microsoft QnA!

Based on the Azure SDK for Python documentation, the begin_classify_document method of the DocumentAnalysisClient class is used for classifying documents using a custom classifier. However, there is no direct parameter in this method to set the classification to be done at the document level instead of page by page. The method signature is as follows:

from azure.core.credentials import AzureKeyCredential from azure.ai.formrecognizer import DocumentAnalysisClient endpoint = os.environ["AZURE_FORM_RECOGNIZER_ENDPOINT"] key = os.environ["AZURE_FORM_RECOGNIZER_KEY"] classifier_id = os.getenv("CLASSIFIER_ID", classifier_id) document_analysis_client = DocumentAnalysisClient( endpoint=endpoint, credential=AzureKeyCredential(key) ) with open(path_to_sample_documents, "rb") as f: poller = document_analysis_client.begin_classify_document( classifier_id, document=f ) result = poller.result() print("----Classified documents----") for doc in result.documents: print( f"Found document of type '{doc.doc_type or 'N/A'}' with a confidence of {doc.confidence} contained on " f"the following pages: {[region.page_number for region in doc.bounding_regions]}" )

In this code snippet, the classification is performed on the whole document, but the SDK does not provide an explicit argument to enforce single classification for the entire document if it inherently processes it page by page

.

If you need a single classification for the entire document and the SDK doesn't support it directly, you might consider a post-processing step where you aggregate the page-level classifications to derive a document-level classification, based on your application's logic.

I hope this helps!

The answer or portions of it may have been assisted by AI Source: ChatGPT Subscription

Kindly mark the answer as Accepted and Upvote in case it helped!

Regards
VasaviLankipalle-MSFT 18,676 Reputation points Moderator

2024-01-06T03:01:59.6666667+00:00
Hello @Roberto Araujo Filho , Thanks for using Microsoft Q&A Platform.

Please note that,

The public preview version of Document Intelligence client libraries default to REST API version 2023-10-31-preview. Starting with the 2023-10-31-preview API, analyzing documents with the custom classification model won't split documents by default.

You need to explicitly set the splitMode property to auto to preserve the behavior from previous releases. The default for splitMode is none. If your input file contains multiple documents, you need to enable splitting by setting the splitMode to auto. https://learn.microsoft.com/en-us/azure/ai-services/document-intelligence/concept-custom-classifier?view=doc-intel-4.0.0

May I know the SDK version and the python version you are working on? Make sure you are using the latest SDK version and python>=3.8.

python -m pip install azure-ai-documentintelligence

This table shows the relationship between SDK versions and supported API service versions: https://github.com/Azure/azure-sdk-for-python/blob/main/sdk/documentintelligence/azure-ai-documentintelligence/samples/sample_classify_document.py

Please follow the documentation to understand how split parameter is set using REST API: https://learn.microsoft.com/en-us/rest/api/aiservices/document-classifiers/classify-document?view=rest-aiservices-2023-10-31-preview&tabs=HTTP#splitmode
Roberto Araujo Filho 115 Reputation points

2024-01-07T01:28:12.6566667+00:00
Thank you very much VasaviLankipalle-MSFT,

The problem was that I was calling an old SDK version (I think it was V3.1 2023-07-31 (GA)):

# old code from azure.ai.formrecognizer import DocumentAnalysisClient document_analysis_client = DocumentAnalysisClient( endpoint=endpoint, credential=AzureKeyCredential(key) ) with open(file, "rb") as f: poller = document_analysis_client.begin_classify_document(classifier_id, document=f) result = poller.result()

So, 'begin_classify_document' method had no argument 'split' (SplitMode).

Today I found the correct way to call V4.0 SDK (2023-10-31 (preview)), but I could´t find documentation about classification for this SDK version, so I didn´t know the correct syntax/arguments to use with 'begin_classify_document' method of this version.

Finally I could find the correct usage by the example you sent: [ https://github.com/Azure/azure-sdk-for-python/blob/main/sdk/documentintelligence/azure-ai-documentintelligence/samples/sample_classify_document.py]

Basicly, changed 'azure.ai.formrecognizer' to 'azure.ai.documentintelligence' and then could set 'split' argument:

# new code from azure.ai.documentintelligence import DocumentIntelligenceClient document_analysis_client = DocumentIntelligenceClient( endpoint=endpoint, credential=AzureKeyCredential(key) ) with open(each_file, "rb") as f: poller = document_analysis_client.begin_classify_document( classifier_id, classify_request=f, split="none", content_type="application/octet-stream" ) result = poller.result()

Thanks again!!!
Roberto Araujo Filho 115 Reputation points

2024-01-07T01:34:25.48+00:00

Thank you for your answer Konstantinos Passadis,

You are right if we talk about the V3.1 SDK. But the new V4.0 SDK supports an explicit argument (split) to enforce single classification for the entire document. I just didn´t know how to use it as I couldn´t find documentation. Take a look at the comments.
Roberto Araujo Filho 115 Reputation points

2024-01-07T01:39:13.6133333+00:00

NOTE: the new code, using the 'azure.ai.documentintelligence' library, takes about 4.5 times longer to classify a document compared to the old code that used 'azure.ai.formrecognizer'...
Konstantinos Passadis 19,586 Reputation points MVP

2024-01-07T03:45:24.83+00:00

Hello @Roberto Araujo Filho !

Thats great !

VasaviLankipalle-MSFT Do you mind to recap the actions and the user will set an answer as Accepted ?

Regards
VasaviLankipalle-MSFT 18,676 Reputation points Moderator

2024-01-07T03:57:50.0466667+00:00

@Roberto Araujo Filho , I have reposted the answer with all these findings that helped you to resolve the issue. Please take time in accepting the answer if you feel helpful. Thanks!
Konstantinos Passadis 19,586 Reputation points MVP

2024-01-07T04:01:06.2966667+00:00

VasaviLankipalle-MSFT , thank you !

this post helped me also !

Accepted answer

0 additional answers

Your answer

Konstantinos Passadis 19,586 Reputation points MVP

2024-01-05T23:32:18.69+00:00

Hello @Roberto Araujo Filho !

Welcome to Microsoft QnA!

Based on the Azure SDK for Python documentation, the begin_classify_document method of the DocumentAnalysisClient class is used for classifying documents using a custom classifier. However, there is no direct parameter in this method to set the classification to be done at the document level instead of page by page. The method signature is as follows:

from azure.core.credentials import AzureKeyCredential from azure.ai.formrecognizer import DocumentAnalysisClient endpoint = os.environ["AZURE_FORM_RECOGNIZER_ENDPOINT"] key = os.environ["AZURE_FORM_RECOGNIZER_KEY"] classifier_id = os.getenv("CLASSIFIER_ID", classifier_id) document_analysis_client = DocumentAnalysisClient( endpoint=endpoint, credential=AzureKeyCredential(key) ) with open(path_to_sample_documents, "rb") as f: poller = document_analysis_client.begin_classify_document( classifier_id, document=f ) result = poller.result() print("----Classified documents----") for doc in result.documents: print( f"Found document of type '{doc.doc_type or 'N/A'}' with a confidence of {doc.confidence} contained on " f"the following pages: {[region.page_number for region in doc.bounding_regions]}" )

In this code snippet, the classification is performed on the whole document, but the SDK does not provide an explicit argument to enforce single classification for the entire document if it inherently processes it page by page

.

If you need a single classification for the entire document and the SDK doesn't support it directly, you might consider a post-processing step where you aggregate the page-level classifications to derive a document-level classification, based on your application's logic.

I hope this helps!

The answer or portions of it may have been assisted by AI Source: ChatGPT Subscription

Kindly mark the answer as Accepted and Upvote in case it helped!

Regards
VasaviLankipalle-MSFT 18,676 Reputation points Moderator

2024-01-06T03:01:59.6666667+00:00

Hello @Roberto Araujo Filho , Thanks for using Microsoft Q&A Platform.

Please note that,

The public preview version of Document Intelligence client libraries default to REST API version 2023-10-31-preview. Starting with the 2023-10-31-preview API, analyzing documents with the custom classification model won't split documents by default.

You need to explicitly set the splitMode property to auto to preserve the behavior from previous releases. The default for splitMode is none. If your input file contains multiple documents, you need to enable splitting by setting the splitMode to auto. https://learn.microsoft.com/en-us/azure/ai-services/document-intelligence/concept-custom-classifier?view=doc-intel-4.0.0

May I know the SDK version and the python version you are working on? Make sure you are using the latest SDK version and python>=3.8.

python -m pip install azure-ai-documentintelligence

This table shows the relationship between SDK versions and supported API service versions: https://github.com/Azure/azure-sdk-for-python/blob/main/sdk/documentintelligence/azure-ai-documentintelligence/samples/sample_classify_document.py

Please follow the documentation to understand how split parameter is set using REST API: https://learn.microsoft.com/en-us/rest/api/aiservices/document-classifiers/classify-document?view=rest-aiservices-2023-10-31-preview&tabs=HTTP#splitmode
Roberto Araujo Filho 115 Reputation points

2024-01-07T01:28:12.6566667+00:00

Thank you very much VasaviLankipalle-MSFT,

The problem was that I was calling an old SDK version (I think it was V3.1 2023-07-31 (GA)):

# old code from azure.ai.formrecognizer import DocumentAnalysisClient document_analysis_client = DocumentAnalysisClient( endpoint=endpoint, credential=AzureKeyCredential(key) ) with open(file, "rb") as f: poller = document_analysis_client.begin_classify_document(classifier_id, document=f) result = poller.result()

So, 'begin_classify_document' method had no argument 'split' (SplitMode).

Today I found the correct way to call V4.0 SDK (2023-10-31 (preview)), but I could´t find documentation about classification for this SDK version, so I didn´t know the correct syntax/arguments to use with 'begin_classify_document' method of this version.

Finally I could find the correct usage by the example you sent: [ https://github.com/Azure/azure-sdk-for-python/blob/main/sdk/documentintelligence/azure-ai-documentintelligence/samples/sample_classify_document.py]

Basicly, changed 'azure.ai.formrecognizer' to 'azure.ai.documentintelligence' and then could set 'split' argument:

# new code from azure.ai.documentintelligence import DocumentIntelligenceClient document_analysis_client = DocumentIntelligenceClient( endpoint=endpoint, credential=AzureKeyCredential(key) ) with open(each_file, "rb") as f: poller = document_analysis_client.begin_classify_document( classifier_id, classify_request=f, split="none", content_type="application/octet-stream" ) result = poller.result()

Thanks again!!!
Roberto Araujo Filho 115 Reputation points

2024-01-07T01:34:25.48+00:00

Thank you for your answer Konstantinos Passadis,

You are right if we talk about the V3.1 SDK. But the new V4.0 SDK supports an explicit argument (split) to enforce single classification for the entire document. I just didn´t know how to use it as I couldn´t find documentation. Take a look at the comments.
Roberto Araujo Filho 115 Reputation points

2024-01-07T01:39:13.6133333+00:00

NOTE: the new code, using the 'azure.ai.documentintelligence' library, takes about 4.5 times longer to classify a document compared to the old code that used 'azure.ai.formrecognizer'...
Konstantinos Passadis 19,586 Reputation points MVP

2024-01-07T03:45:24.83+00:00

Hello @Roberto Araujo Filho !

Thats great !

VasaviLankipalle-MSFT Do you mind to recap the actions and the user will set an answer as Accepted ?

Regards
VasaviLankipalle-MSFT 18,676 Reputation points Moderator

2024-01-07T03:57:50.0466667+00:00

@Roberto Araujo Filho , I have reposted the answer with all these findings that helped you to resolve the issue. Please take time in accepting the answer if you feel helpful. Thanks!
Konstantinos Passadis 19,586 Reputation points MVP

2024-01-07T04:01:06.2966667+00:00

VasaviLankipalle-MSFT , thank you !

this post helped me also !

Answer 1

Hello @Roberto Araujo Filho , I'm glad that you were able to resolve your issue and thank you for posting your solution so that others experiencing the same thing can easily reference this! Since the Microsoft Q&A community has a policy that "The question author cannot accept their own answer. They can only accept answers by others ", I'll repost your solution in case you'd like to "Accept " the answer.

Issue: Azure Custom Classification Model - How to set "split mode" with Python?

Solution:

The public preview version of Document Intelligence client libraries default to REST API version 2023-10-31-preview. Starting with the 2023-10-31-preview API, analyzing documents with the custom classification model won't split documents by default. You need to explicitly set the splitMode property to auto to preserve the behavior from previous releases. The default for splitMode is none. If your input file contains multiple documents, you need to enable splitting by setting the splitMode to auto. https://learn.microsoft.com/en-us/azure/ai-services/document-intelligence/concept-custom-classifier?view=doc-intel-4.0.0

Make sure you are using the latest SDK version and python>=3.8.

python -m pip install azure-ai-documentintelligence

This table shows the relationship between SDK versions and supported API service versions: Here is the sample code: https://github.com/Azure/azure-sdk-for-python/blob/main/sdk/documentintelligence/azure-ai-documentintelligence/samples/sample_classify_document.py User's image

The sample SDK code to set 'split' mode using new SDK V4.0 (1.0.0b1 (preview)):

# new code
from azure.ai.documentintelligence import DocumentIntelligenceClient

document_analysis_client = DocumentIntelligenceClient(
    endpoint=endpoint, credential=AzureKeyCredential(key)
)

with open(each_file, "rb") as f:
    poller = document_analysis_client.begin_classify_document(
		classifier_id, classify_request=f, split="none", content_type="application/octet-stream"
	)
result = poller.result()

If you have any other questions or are still running into more issues, please let me know.

Thank you again for your time and patience throughout this issue.

Regards,
Vasavi

Please remember to "Accept Answer" if any answer/reply helped, so that others in the community facing similar issues can easily find the solution.

Share via

Azure Custom Classification Model - How set "split mode" with Python?

0 additional answers

Your answer