You can run tests against the project file to find entries in the training documents that are probably not correct. The generated report highlights any errors that were found and allows quick navigation to the documents that are failing.
The following errors can occur.
-
Unable to find a given field value on a document. This means that the
value will not be used for learning. This is not a problem if it happens in a
few training documents but if no values can be found for a field within an
entire class, learning that field would fail.
Review the document and make sure that the value is correct and is appropriate for the field type.
-
Unable to extract a value for a given field when extracting the training
document. This indicates that the training set is not good
enough yet.
Consider adding more training documents with further examples for this field.
-
Document classified into a different class from the one that it was stored
in.
Review the document and check if it really belongs in this class. Consider removing it from the training set if you still think that the assigned class is correct - it might just not be a good example.
- As a more general issue, classification of all documents of one class into a
different class. This usually means that the two classes are so similar that
they cannot be differentiated.
Either merge the two classes into one, or add additional training documents that emphasize the differences between the classes.
To test a project file, complete the following steps.
-
On the class viewer area of a project, click the Test this
project icon
.
- Click on an incident to view the training document with the errors highlighted.
- To return to the report, click on the relevant project folder.