Decipher is not reading Invoice number from invoices

[charanReddy]

Decipher is not reading Invoice number from invoices

We have trained 20-25 different type of invoices they used work as expected upto last week but suddenly from Monday decipher is not reading invoice field for previously trained invoices
Remaining fields are working as expected. we haven't changed any thing in invoice field's of DFD.

Please let me know if you need any other details

​
#decipher #deciphertectips #unabletoreadinvoicenumber ​​​​

------------------------------
charan Reddy
developer
accenture
Asia/Kolkata
------------------------------

[HarpreetKaur]

Hi Charan,

Have you changed anything else in the structure of the DFD? Added/Deleted fields or tables or changed their ID by any chance?
Any of these changes have the potential to reset the underlying ML model associated with that DFD.
http://bpdocs.blueprism.com/decipher/user-guide/machine-learning.htm

Regards
Harpreet

------------------------------
Harpreet Kaur Product Consultant
------------------------------

[charanReddy]

We have not added/deleted any new fields or tables or changed any ids.

We have added sample headers & flags to some other fields but no change is done to invoice number field.

Does adding headers or flags to existing fields cause this issue?

------------------------------
charan Reddy
developer
accenture
Asia/Kolkata
------------------------------

[BenLyons]

Hi Charan,

This is a difficult query to answer without being able to see it all first hand, but this is the best guidance I can give.

Decipher should not behave differently from one day to the next if no changes have been made. There are a number of changes that can impact this performance, I'm not sure I could give you a complete list.

I recommend reviewing all changes that were made between the last successful document and the first unsuccessful document. It may be that the respective changes have affected the training data and you will need to upload more documents for training.

If absolutely no changes have been made, then I recommend raising a support ticket for further investigation.

Thanks

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

[IsmoLehtiniemi]

Do you have Machine Learning enabled for the DFD? 

From our experience I could comment the the location & header etc. based recognition logics seem to work some times worse more we run documents through Decipher, when ML is enabled.

So at least in theory when you use the ML option, and when your environment hits the magical 1000 (or whatever count is set as the ML original dataset) processed documents, your result might get worse because the ML is overwriting the original location & header etc. rules, which worked fine previously.

------------------------------
Ismo Lehtiniemi
RPA Architect & Solution Expert
Amcor
Europe/Warsaw
------------------------------

[charanReddy]

we have changed the flag for another field (i.e. not for invoice number field). can adding or removing a flag or a header for another filed,  be a reason for this problem?

After noticing this issue we reverted the change and tried to retrain invoices but we are still facing the issue. What steps do you suggest to us to make decipher identify Invoice numbers ?

------------------------------
charan Reddy
developer
accenture
Asia/Kolkata
------------------------------

[BenLyons]

Hi Charan,

I can see no reason for this to happen without further investigation.

The quickest way to get back to normal would be to delete your training data and start training again. This will delete training for all your documents, so be mindful of this before deleting.

Otherwise you will need to reattempt your original training, but it may take a little longer as something unexpected has happened.

Also, try considering other factors which may make locating the Invoice Number more consistent e.g. Regex, Sample Headers, Validation Lists.

Thanks

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