Can anyone tell me that how we can look at the backend code of the out of packages activities especially the packages shown in screenshot? I have to make a custom model like this. I have also checked solutions and different tabs in uipath marketplace but did not find any.
Any type of help will be grateful.
so u need to know how to make a custom model like this and upload to ai center
So check out the video
@NIVED_NAMBIAR Nopes, I Know the Custom activities deployment in AI Fabric.
I want to make the other custom activity myself. So, Should I know the code of custom activities like how they are actually made?
the custom models are based on python
i think going through the video would help you a lot in understanding
@NIVED_NAMBIAR can I have that python code of invoices purchase order custom activities??
It would be a great help.
i am not sure about that
you can have a look here
@NIVED_NAMBIAR it is not what I am looking for.
What I am asking is code of this machine learning model.
Do you have any idea of it?
This code is not public so you won’t be able to find it.
Our company is working on custom models of medical forms. They would be similar to invoices machine learning model extractor. We will be glad if you provide it in open source.
More people will make custom models and use AI Fabric.
@nashrahkhan have you tried using generic document understanding model and just retraining it? It’s been built to adapt on different document type.
We can’t open sourced this model at the moment as it has a lot of IP but we can work with customers on building new similar models.
Including @alexcabuz working on this topic.
We are interested to help provide Out of the Box models for the most common types of documents. What kinds of medical forms do you have in mind? Like the CMS 1500 type forms? Any other types of forms?
@alexcabuz Thanks for responding.
Well we have 10 folders with different types of medical documents in it. The max size limit in some folders are 500 and minimum is 5. Attaching here examples of some forms UI.
For Form 1 we also created MLPackge models as per our understanding but it is not giving us results as we made it non retrainable. To make it trainable we are unable to do so because of lack of custom activities code. According to DS team, instead of Google OCR in their package they will make their own OCR from scratch which is time consuming and taking 2 months. Can you kindly guide regarding this that how we can make custom models and used it in training our models.
So what I am seeing here, are a few separate topics:
can you get the ML package source code? No. It is UiPath proprietary. You CAN however use it to train UiPath’s Document Understanding ML model with your files. it has the advantage that it comes with the ML Extractor activity as well, so nothing else to do from your side, just label data and train your model. It is also retrainable.
Custom OCR engine activity - yes, you can build your own. See this repo Document-Processing-Code-Samples/Samples/SampleActivities/Basic at master · UiPath/Document-Processing-Code-Samples · GitHub along with this public contract documentation here https://docs.uipath.com/document-understanding/docs/about-the-ocr-contracts on how to build your own OCR engine. Of course, it would not be available to be used directly in Data Manager in Ai Center, BUT you can collect your training data by building a simple workflow that does Digitize (with your OCR engine), then Validation Station (so human can perform the labeling), then Train Extractors Scope with a ML Extractor Trainer inside configured properly.
Custom Extractor with a Custom ML Model hosted in AI Center, but NOT starting from the DU ML model: the simplest approach would be to publish whatever custom ML model you wish into AI Center, use the ML Skill activities directly from AI Center to call your model, then build a custom Extractor (using the same github repo example along with the public document processing contracts package https://docs.uipath.com/document-understanding/docs/about-the-document-processing-contracts ) to MAP your raw ML model output to exactly what is needed from an extractor.
I hope this helps