I know this question may be hard to answer, but how many training documents build a well-trained model?
For example, for check processing.
I have trained 55, and the accuracy is okay. Problem is, there are many different slightly different formats of checks. When I have time, I will try to train on 100 to see how that improves.
Hi Niko,
Giving an answer that will apply in all cases is uneasy.
You should first train a model with 5 samples and check what is the accuracy.
If it is not good, try to add more samples and leverage the supervised feature to give hints to the model.