Hi. is it possible that existing ML model can be trained addtionally in the following scenarios? and how? 1) when accurracy of part of the existing lables isn’t good enough, the user wants to have the ML model learn with more documents only for those labels 2) when ML model requires the new labels to be trained, user wants to train the model with more documents with new labels. (assume that there’s no previous dataset remained)
@hongjoo.choi Yes! you can retrain an existing ML model in both scenarios you described
Thank you @Darshan_Sable . Then, is it possible to train the ML Model using new dataset witih only part of those lables (let’s say 3 out of 10 labels) without degrading the accuracy of the rest lables (e.g., 7 lables in this case)?
@hongjoo.choi Yes each label has its own accuracy. You can add new labels. But each label should be annotated at least 10 times in Data Manager. Higher the number relates to more accuracy
@Darshan_Sable Ok. Just for clarification, regardless of whether the labels are newly added or existing ones, all the existing lables + new ones should be annotated to the new dataset. Right?
@hongjoo.choi
Yes, that’s correct! Regardless of whether the labels are newly added or existing ones, all labels both old and new should be annotated to the new dataset. This ensures that the ML model is trained comprehensively and recognizes both the previous and newly introduced labels.
Ideally you should be annotating all..and if pther fields are properly identified previously with different docs then it should replicate the same
And the new fields added will also be trained..but training happens on full set with full fields
Cheers
This topic was automatically closed 3 days after the last reply. New replies are no longer allowed.