Batch stuck in Waiting for Class Training

[FredericTaes1]

Hi, 

I'm trying to test the classification capabilities of Decipher.
I prepared a batch of 53 documents that were sent to Decipher (as a training batch). After the OCR step, the batch enters the status "Waiting for Class Training". However, it has been in this state for more than an hour now, and I see no progress being made. 
Is it a matter of having even more patience, or are there certain steps to be taken to actually start the classification training ?

------------------------------
Frederic Taes
RPA Consultant
RoboRana
------------------------------

[BenLyons]

Hi Shweta,

The most likely cause of that is that you have already trained a batch on the classification model.

When training the model, you need to have all batches at the stage "Awaiting classification training" before enabling the classification model for training.

Thanks

------------------------------
Ben Lyons
Senior Product Specialist - Decipher
Blue Prism
UK based
------------------------------

[ShwetaYadav]

Alright Thanks Ben. I got your point. 

In addition to this let say I have trained a document to capture specified fields and set the periodic training & training size for small set of 10 docs in the capture model.

Still when I push the same document at the 11th time - why do I have to re-indicate the field value that was already trained?? I am confused, does it require more training if so then how much training is required in that case or there is some other way to handle this?

Thanks
------------------------------
Shweta Yadav
------------------------------

[BenLyons]

Hi Shweta,

10 is much too small of a number to train an effective ML capture model, and very likely unnecessary at this stage. Decipher has a primary ML engine that's always on and will probably do much better with learning on such a small number of documents.

I recommend reading through our best practice guidance to see how you can get the best out of Decipher.

Thanks

------------------------------
Ben Lyons
Senior Product Specialist - Decipher
Blue Prism
UK based
------------------------------

[ShwetaYadav]

Hi Ben,

Even After Disabling the ML model, I tried to simply push the document and verify the data, but every time i do that there is something wrong with any of the field specified in the DFD . I try to edit DFD restart the batch to re capture the data, but no luck.
Can you please advice.

Thanks

------------------------------
Shweta Yadav
------------------------------

[BenLyons]

Hi Shweta,

You may need to delete your training data, as it will be impacted by all your previous training.

But again, I would recommend you read the best practice guidance first as it will help confirm if the training to date is in keeping with the recommended standards.

Thanks

------------------------------
Ben Lyons
Senior Product Specialist - Decipher
Blue Prism
UK based
------------------------------

[ShwetaYadav]

Hi Ben,

I deleted the training data and started from the fresh, also referred the same steps as mentioned in the best practices doc.

1. Like you suggested to focus on the DFD configuration adjustments and train the batch- restart to see if it captures what is expected, also added few regex patterns and tried to indicate manually on the field region which has to be extracted and submit the batch.

2. Again in the next push it appears to be as in the previous batch. I keep on training but its not ready to learn 😞

3. Also I tried to implement ML by enabling the same in the document type. (Same is the case with this too, changes not reflecting as expected. )

Thanks


------------------------------
Shweta Yadav
------------------------------

[BenLyons]

Hi Shweta,

It sounds like you're doing everything right, but it's very hard to get a true feel for someone else's documents on this channel.

You may be entitled to some Expert Connect sessions, these can be requested via the customer support team. This would give you a chance to work directly with a Decipher expert on your use case.

Thanks

------------------------------
Ben Lyons
Senior Product Specialist - Decipher
Blue Prism
UK based
------------------------------

[ShwetaYadav]

Hi Ben,

I can now extract the data after following the suggested steps by the expert.
Now I am stuck on identifying of Document Type, my batch contains 5 types of document how to retrieve the document type from the queue items.

Can you please suggest.

Thanks

------------------------------
Shweta Yadav
------------------------------

[BenLyons]

Hi Shweta,

If your classification model isn't working as you'd hoped, you have 2 options I can think of to improve.

If you can gather more document samples to train a new model with, ideally at least 50 of each document type. A more comprehensive model should produce better results.

Alternatively, if the confidence is close, you can reduce the threshold by document type (see the Document Type settings). For example, the classification confidence for document type A is above 80%. Change the threshold just for that document type.

If your document types are very similar, you may still struggle. For example if all 5 types are different invoice formats, they will likely contain a lot the same key fields e.g. invoice number, invoice total.

Thanks

------------------------------
Ben Lyons
Senior Product Specialist - Decipher
Blue Prism
UK based
------------------------------

[ShwetaYadav]

Hey Ben,

Hope  you are doing good!!

I think I framed the question in a wrong way, Actually I am able to identify the document type in the Decipher tool.
But when I retrieve the extracted batch details using the Queue items, I might have same or different set of document types in a batch, what is the process to get the document type for the extracted data for each doc?

Mostly useful when trying to consolidate the batch data in a single report.

Thanks

------------------------------
Shweta Yadav
------------------------------