I created document understanding package, trained with sample files using pipeline then created skill.
When I ran the skill in studio, It managed to fetch the desired data for the documents I provided but one particular not one particular field in one particular document. Then I added the same document in dataset and trained the package and selected the trained version and deployed. But still DU failed to label that field.
Please check below image, which is trained in in AI Center.
If you check the question…it clearly says you added the field only for one particular document…not for all…so it is not evident to understand…as per this we see no other document has that field…
Can you confirm if all 10 have that field and you labelled in all 10 coduments?
Sorry, If my question created confusion. I trained 10 document of same format all the fields labeled in all 10 documents. The last document which is in screenshot, I am trying to process the same document with funding amount as 470. But it was trained with value 480. If I process with funding amount as 460. Bot can fetch the data, Only problem with funding amount as 470.
Generally, we would train with more than 10 documents, even if 10 documents are a minimum we go with with 20-25 of the same format and then Train the Model with this dataset.
Also, do remember that each stage of Re-Training, Train the Base Version with all the Set Of Labelled data.
Let us know if you are able to label for 20-25 documents of the same format, we can then check if it does not correspond to the field mapping.
Also, not a very complete info is given on the fields or on How the Training was done? How many versions of the Model were already created ?
Training done via Train pipeline using dataset exported from Data labeling session. I will try feed 25 docs and let you know the outcome. It was trained only on based version 0. ML skill deployed using Version 1.
Alright. Let’s keep the base version as 0 and move forward with the Pipeline Training when also additional training dataset is added, Train the whole dataset and not just the additional data that was added.
Let us know once you were able to use the ML Skill and if there are improvements observed.
Hi, try create a new regular field then export with the name please check the (ALL LABELED).
Then go to pipeline use the minor version and use in studio the machine learning extractor.
Wait the package is successfull in status in pipeline then use only if is successful.
Then create new evaluation to see the score.
sonaliaggarwal47
Aditya_Nalawade11
Lynn_Song
ashokkarale
Anil_G
arivu96
sharazkm32