Hi Krishna,
It's quite difficult to fully appreciate the steps you have taken without seeing it, but I'll offer as much as I know on how you can explore a resolution.
- When training for a new document (with ML model disabled), I personally found that verifying a document, then making DFD changes would increase the number of documents required to successfully train. The best process I found was to create the DFD and upload 1 document, check how well it works in the verification screen. Then if there are changes required, return the document without saving or submitted and making the needed changes. In batches, restart your uploaded batch at the capture stage before going to verification. Following this may help produce better results.
- The only way to reset 100% of your training is to delete your training data, deleting your DFD will not remove any training. This is not recommended in 1.1 as you will delete training for all documents and would be unable to restore it. As such you will need to train a higher number of documents to resolve any updates. Once 1.2 is released this process can be easier where training can be exported/imported. This is due very shortly.
- In addition this could be a bug (unable to confirm without detailed investigation), of which a number have been resolved in 1.2. So I apologise for this experience and hope that upon the release of 1.2 you are able upgrade to get the latest improvements.
Thanks
Ben
Ben Lyons
Principal Product Specialist - Decipher
SS&C Blue Prism
UK based