Empty JSON result using "MLSkill" activity (Custom Named Entity Recognition - AI Center)

Help AI Center
, , ,
1

How to find out why my ML skill output is empty?

Below please find what I have done

In Label Studio
-Create project in label-studio

image
image895×702 23.7 KB

In AI Center
-Upload the exported text file from label studio to “Dataset”

image
image790×699 16.9 KB

-Create Pipelines with below parameters
-model.epochs: 5
-dataset.input_format: label_studio
-removed the other 2 parameters

In UiPath Studio
-Create MLSkill activity
-Result as below
image

Main question: How to identify the problem? (No idea how to debug right now)

Other questions:
-What is the suggested dataset sample size? I am currently using 42 set of text, labeling 2 fields for each dataset
-Any tips or good resources on learning how to do custom NER using UiPath?

Thanks!

2

You can try training the model with more epochs (maybe 50) and give a try to see how the results look. The output is empty, because the model was not able to extract those fields from the input text. Sharing some of the documentation links below for NER model :
https://docs.uipath.com/ai-center/automation-cloud/latest/user-guide/custom-named-entity-recognition

https://docs.uipath.com/ai-center/standalone/2022.4/user-guide/extracting-chemicals-from-research-paper-by-category

3

Hi @Anas-p-v
Thanks for the reply
Used 50 epochs, result in below error
Kubernetes operation failed to create deployment

May I ask how to solve this

4

Are you using GPU for the skill deployment, if so please try on CPU once.

5

@Anas-p-v
Thanks for following up and the quick reply!
Tried on CPU, result in the same error.