Training my custom ml skill and ml package for specific document type

Help Document Understanding
, , ,
1

I have my custom ml skill and ml package for my document type. This type includes different fields, such as tables and regular fields. I have a problem with varying values in table columns. I train the ml package on various documents, but if there are minor changes in the document using my ml skill, I have unexpected and wrong results. Likewise, I wonder if there is some way to train specific fields in the document and train the process to avoid this error in my next documents. I try to investigate Train Extractors Scope and Machine Learning Extractor Trainer but don’t find something helpful.

2

@Stepan_Zhemevko

Is the column name itself changing or only values?

can you please show how you trained…and is it happening only for column fields or for regular fields as well?

cheers

3

Thanks for Reply!
The column name isn`t changing, only values can be different and Yes it happens only for column fields. About trained I mean that I added to the Document Manager a new batch with documents that have such problems, after that, I exported bathes to the AI Center and used the Train Pipeline train ML Package. Then based on a new version of ML Package I create ML Skill and use it in Studio in my workflow.

4

@Stepan_Zhemevko

Ideally it should be working in that scenario …can you please show some screenshots on how you are doing the trainign and how the labelng is done

cheers

6

Unfortunately, I cant show you the documents and labeling itself. Toking about training I simply create a new ML Package version in AI Center with new documents. I think the main question for me will be how to use Train Extractors Scope and Machine Learning Extractor Trainer activities to improve my workflow.
Thank`s!

7

@Stepan_Zhemevko

When a new document comes in from validation stage then you might need to include the training part in it…for detailed activities.check the DU Template

cheers

8

Thank you! I will try to investigate this more

1 Like
9

Hi @Stepan_Zhemevko

Ideally you would require 10 Documents for single fields for training purpose then Creating pipeline, after success and create skill use it in ML skill activity, result should be good then.

Regards