How to import and retrain the dataset in document understanding without using data manager or orchestrator for create or import dataset

Help Document Understanding
, ,
1

I’m working on the docuemnt understanding to parse the different format of Invoice pdfs,

  1. So I want to retrain the UiPath invoice model.
  2. Help me to retrain the model by import and retrain the model without using the data manager or Orchestrator for dataset. I want to import it locally through the UiPath Studio.
    @ppr @nisargkadam23 @RAKESH_KUMAR_BEHERA @Vibhor.Shrivastava @mukeshkala @sandeep13
2

Hi @Ajay_Kumar10 ,

Maybe you require to auto-export the validated data back to the Dataset, if so, you could check the below post :

If the above doesn’t clarify, you could provide us some more info on what is the actual requirement.

3

Hi @supermanPunch,
I want to retrain the ML model, but when I create the datasets and import human validation station training data through the machine learning extractor trainer. after that, I create and run ML pipeline so it is failing.

image
image1418×777 37 KB

image
image1907×885 55.3 KB

4

@Ajay_Kumar10 ,

What was the configuration done in pipeline ? What was the dataset selected ?

We would require to Perform the Export feature in Document Manager (Or Schedule this), so that the fine tune folder data gets combined with the previous data as well and gets stored in the export folder. This way we would require to select the export folder as the dataset for the pipeline and set auto-retraining parameter to true.

The Second point in the Topic suggested in the previous post mentions that :

image
image797×264 14.1 KB

1 Like
5

@Ajay_Kumar10 -
We have two option for retraining.

  1. Manual:
    Once the Document are validated in Action Center, then keep the document in local/shared drive folder, it will create a three folders. Zip all the three folder and import in Data Manager then export it to dataset and start the pipeline.
  2. Auto Retraining -
    You have to schedule this in Data Manager & Pipeline.
    Recommended is manual so that you will have the track of retraining data.

Let me know if you have any question