When creating a training dataset, what is the recommended number of samples for the Classification fields?
When creating a training dataset, what is the recommended number of samples for the Classification fields?
When creating a training dataset, the recommended number of samples for Classification fields is 10-20 document samples from each class. Having a sufficient number of samples ensures that the model can learn effectively from a variety of examples, improving its accuracy and robustness.
I think the answer is not correct. It should be 10-20
Classification fields (currency) Classification fields generally require at least 10-20 samples from each class. The recommended range for dataset size is based on the information provided in the Calculator tab. For simpler scenarios with few regular fields and clear document layouts, good results may be obtained with datasets in low orange range. For complex scenarios, especially involving complex tables with many columns, good results may require datasets in high orange or even green range. https://docs.uipath.com/document-understanding/automation-cloud/latest/classic-user-guide/training-high-performing-models
Regular fields: 20-50 samples per field Column fields: 50-200 samples per field Classification fields: 10-20 samples per class So the answer is B
B should be the correct answer. "Classification fields generally require at least 10-20 samples from each class." source: https://docs.uipath.com/document-understanding/automation-cloud/latest/classic-user-guide/training-high-performing-models
B is correct