topic RE: Decipher results not matching training in Product Forum

Decipher results not matching training

JamesKarm — Fri, 10 Sep 2021 18:26:00 GMT

Hi,
Has anyone had issues training decipher on multiple client invoices? That tend to have slight variations?
In training Decipher it seemed to be getting each batch of client's correct. or would learn to be correct.
Then in a random test with the same client batches but new invoices the results afterward were far less than expected.

It is almost as if the training of the next set of client invoices caused it to confuse its previous learnings on others.

Does every new client require a fresh DFD/Batch etc?

Sorry I realize this probably doesn't have a simple answer.

------------------------------
James Karm
developer
new employee of rpai
America/Toronto
------------------------------

RE: Decipher results not matching training

KrishnaElapavul — Sat, 11 Sep 2021 07:00:00 GMT

Hi James,
I too got in to this situation... then I applied below solution and achieved better solution.

No need of different DFD/Batch type for different clients
create classification model training for Invoice
Create 2 different document types and map the classification accordingly ( CDT1 and CDT2)
Link both Document types to same Batch type ( BT1)
Decipher would able to map the pushed invoice to specific DT1/DT2 ( Can notice in Class verification and data verification)
You may see better results.

------------------------------
Krishna Elapavuluri RPA Solution Lead
TEchnology Consultant
DXC.technology
Asia/Kolkata
------------------------------

RE: Decipher results not matching training

Ben.Lyons1 — Mon, 13 Sep 2021 07:08:00 GMT

Hi James,

Generally speaking it should not be necessary to create an additional DFD/Document Type to train additional document formats, though there may be occasions where this is not true.

Do you have ML enabled yet?

The reason I ask is that the native training model (pre-ML capture model) is more closely linked to a document format, so there should be no interaction between training different formats. If Decipher is yet to validate a new format, it will only use the DFD configuration to read it.

Whereas the ML capture model is more general in its application, so a trained model will inform every document format uploaded, whether it has seen it before or not. If the model is not adequately trained, this could cause the behaviour noted above.

Thanks

------------------------------
Ben Lyons
Product Consultant
Blue Prism
UK
------------------------------