30-09-24 07:02 PM
Will the Rule-based training data that was exported from Decipher IDP 2.2 as a TrainingData_2024.td file work as expected when imported into Decipher IDP 2.3 ?
Answered! Go to Answer.
03-10-24 08:06 AM
Hi,
For the majority of cases migrating from 2.2 to 2.3 resulted in consistent accuracy, however some customers reported a similar drop in success as mentioned by Joseph.
This was rectified in last month's release 2.3.2 and should resolve the change in performance experienced.
Customers on 2.2 and looking to upgrade should consider 2.3.2 as their upgrade path.
Thanks
01-10-24 01:44 PM
Hi JD_CPU.
The training data from 2.2 will work as expected in 2.3.
jack
01-10-24 02:25 PM
Thank you, jack.
Unfortunately, that has not been my experience. Created a brand new Decipher IDP 2.3 environment, imported a rules-based .td file that was providing 99% accuracy on Decipher IDP 2.2. After manually data verifying 30+ documents into 2.3 we are still seeing only about 60% accuracy. The only difference between the 2.2 and 2.3 environment is that the 2.3 version is currently only provisioned with 8GB of RAM while the 2.2 version runs on 16GB of RAM.
Is there any chance that the difference in RAM could account for the poor results or was there some change in version 2.3 that causes the 2.2 training to no longer yield the same results as we see in 2.2?
There was a previously known issue here: https://docs.blueprism.com/bundle/decipher-idp-2-3/page/release-notes/rn-2.2.htm Decipher IDP 2.2: March 2023 Reference DEC-774 "An issue has been corrected where training data loaded into a Decipher IDP system which had been upgraded from 1.2 to 2.1 did not function as expected. Now, training data imported into upgraded systems works correctly."
My current issue is somewhat different in that this is NOT a IDP system that has been upgraded. My issue is a brand new 2.3 environment does NOT yield the same results as the 2.2 environment from the same rules-base .td file
01-10-24 02:42 PM
Hi Joseph.
That is interesting. I have not heard this from any of my other clients. There is nothing in the upgrade notes about there being an issue going from 2.2 to 2.3 and I have done this myself.
I don't see how changing the amount of RAM would make a difference in training accuracy.
I suggest opening a ticket with support as they may be able to provide some insight into this issue.
jack
03-10-24 08:06 AM
Hi,
For the majority of cases migrating from 2.2 to 2.3 resulted in consistent accuracy, however some customers reported a similar drop in success as mentioned by Joseph.
This was rectified in last month's release 2.3.2 and should resolve the change in performance experienced.
Customers on 2.2 and looking to upgrade should consider 2.3.2 as their upgrade path.
Thanks
03-10-24 12:38 PM
That's great news! Thank you Jack and Ben for the prompt reply.
Much Appreciate,
JD
03-10-24 01:30 PM
Hi Ben.
That is curious. Thanks for sharing this information.
Cheers
jack