How to use Ui Path Cloud Ai Fabric and Train Extractor Scope Properly?

Good Afternoon All,

I have been struggling finding out information about how to train the current Ai models inside of Ai Fabric’s Document Understanding. I currently have community cloud edition with an enterprise trial active.

I am also using a train extractor scope in my workflow containing the newly added machine learning extractor trainer. I have the Trainer pointed to my skill in Ai Fabric and a folder on my desktop.

My questions are this, how do I achieve actual training of the model? What do I need to do to the folder on my desktop to be able to upload and train the skill on Ai Fabric, since uploading that folder alone throws errors? Does it require interaction with Data Manager? If so are we able to access and use Data Manager with a cloud account and how so?

I know that this is broad range of questions, I have been searching for a way to train the provided Ai models but since the extractor trainer is so new I haven’t been able to find much info about it. Thank you for any help you can provide!

Hi @JosephNehl

Does this article help here a bit? (the last type of import described):
https://docs.uipath.com/ai-fabric/docs/using-data-manager

1 Like

Hi @JosephNehl, thank you for your interest in the Document Understanding product!

Your question might be answered mostly in this section: https://docs.uipath.com/ai-fabric/docs/using-data-manager#machine-learning-extractor-trainer-dataset-import-preview-feature

So the Trainer activity exports data into a folder inside of which it creates 3 subfolders (Documents, Metadata and Predictions) each with an equal number of files. Then you can zip up that folder and import it into a DM instance, but make sure DM has the latest version. Currently the latest version is only available as a standalone docker container which you can run on your Windows 10 laptop, or perhaps on a Linux machine or VM.

After you import it, you export it, and then again unzip the file, and the resulting folder you can upload to AI Fabric Cloud. Another heads up, the ML Packages able to train on these datasets are about to be released in the next couple of weeks (Sept 21 I believe). At that date Data Manager will also be available in Private Preview in AIF Cloud too. So basically what you can do until that time is to run documents through Validation Station, export them using Trainer activity, import them to DM, export them, and prepare for when the next update to AI Fabric Cloud is available.

The training capability is available however in the AI Fabric on prem version 2020.4, which has been patched to include this capability. So if you have access to AI Fabric on prem 2020.4, and an on prem standalone Data Manager container, then you can retrain ML models. If you are using AI Fabric Cloud, then you need to wait another couple of weeks.

The reason detailed instructions are hard to find is because we are waiting for the final pieces of the puzzle to be in place before making the announcement. So you are ahead of the curve!

Hope this helps, and let me know if you run into any issues.

Best,
Alex.

5 Likes

Thank you so much for your detailed explanation!

1 Like

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