Need some details about the Azure Custom Classifier

Question

Hi all, So I have been using up azure custom classifier for a while now and i have few doubts regarding a bunch of things

1. Training the model using API - Have been uploading documents using console but uploading and labelling bulk documents via console (UI) is very painful and iterative training is not available as well, meaning every time it trains everything that there is rather then the new uploaded docs alone, i tried the available methods to train via api from the docs both via azureBlobSource and azureBlobFileListSource, but it fails in every method for me, facing issues like training data not available at path even if files are there and invalid source content issue with jsonl files even tho everything seems to be correct, can someone help with this if someone has already done this?
2. What would be the cost for training the model every time, I only see that it costs $3 per 1000 pages and assuming this is while sending requests across for classification after training? i also see that template model training is free while neural model training is $3 per hour after free hours, can this be explained better and also about the OCR costs if any
3. Also is there any reference material where I can find on an outline how this model works? again the same template vs nerual model, layout vs actual text in document pages, what happens underneath

Thanks and Regards

Barath G

Answer 1

Hi Barath G,
Greetings & Welcome to the Microsoft Q&A forum! Thank you for sharing your query.

Training the model via API can indeed be tricky. Here are a few tips to help you troubleshoot:

Double-check that the paths to your training data are correct and that your JSONL files are properly formatted. Even small errors can cause issues.

Azure Custom Classifier now supports incremental training, which allows you to add new samples to existing classes or add new classes without retraining everything.

If you're facing issues like "training data not available at path" or "invalid source content," ensure that your Azure Blob Storage paths are correct and that your JSONL files match the required schema.

Training Costs:

Custom Template Model: Training is free.

Custom Neural Model: Free for the first 10 hours. After that, it costs $3 per hour.

Classification Costs:

$3 per 1,000 pages for classification requests after training.

OCR Costs: There are no additional OCR costs mentioned specifically for the Azure Custom Classifier, but it's always good to check the latest pricing details on the Azure pricing page. For a detailed outline of how the Azure Custom Classifier works, including the differences between template and neural models, and how layout vs. actual text is handled, you can refer to the following resources:

Build and Train a Custom Classifier.

Custom Classification Model.

I hope this information helps. Thank you!