Retrain the ML Skill from Action Centre

I have trained my custom ML Skill for Invoices by retraining Out of Boxes Packages, Document Understanding for Invoices. So I have deployed my ML Skill and currently in monitoring how well the performance of the ML Skill. Now, I have implemented the Validation Action to Action Centre. After validating the documents in Action Centre,

  1. how can I use the validated data to retrain my ML Skill?
  2. The end to end process (after validation of document until the deployed new ML Skill)

Hi @aqiff

@Lahiru.Fernando has a detailed video tutorial on this:
Part 1: Training UiPath Document Understanding ML Models - Data Manager - Part 1 | RPA - YouTube
Part 2: Training UiPath Document Understanding ML Models - Data Manager - Part 2 | RPA - YouTube

Skip to the part you are facing issues with. In case you have questions, post on this thread again and the community can help you further.

1 Like

Yes, thank you for this. This is what I can’t connect between the two dots. Since the validation action for the document is being sent to Action Centre. So human validated it on Action Centre. So, how can I get the output of validated result from Action Centre, to be the input, Human Validated Data in Train Extraction Scope

Hi aqiff,

Its not totally connected up, which is actually a good thing.
I think the name of the activity can be misleading as it is called ‘Train Extractors’.

What it is actually doing is adding data to your dataset, you should be familiar with this if you say you have already retrained the skill, you will have used a dataset to train it on.
The dataset will be populated with data from each human validated action.

At a later date, as you choose, you can train the ML skill again on this dataset and have it continue its learning based on the human validated data.


I just saw the below post and he accurately points out that you shouldn’t pass extracted data immediately back to the dataset as it is not human validated. You need to use ‘Present Validation Station’ to get a human to validate the data and use the result of that in the Train Extractors scope.

If I understand your question correctly, you are concerned that you’re using the same extractedResults for training, which is no good because it doesn’t contain the values that were updated/corrected during the manual validation in Action Centre.

And that’s accurate, it won’t help your training set at all. But you’re missing an important step, according to the image you’ve posted here.
The Create Document Validation Action is a persistence-supported activity, which sends an action to the Action Centre, to be completed by a human.
This needs to be followed by a Wait for Document Validation Action and Resume activity. This activity automatically makes the job wait in a “Suspended” state and resumes as soon as an action is completed in Action Centre.

As you can see from the image above, the latter activity returns a validatedExtractionResults object which contains the user-corrected data.
This is the object of type ExtractionResults that you want to use in your Train Extractors Scope.

To use the code above, create a project with this template:

Now I understand better ! Thank you so much !

1 Like

This topic was automatically closed 3 days after the last reply. New replies are no longer allowed.