If you want quick and good generic extraction results with a relatively small training set, the highly optimizable Trainable Evaluator, is recommended. This evaluator is used to compare alternatives from other locators to determine which of those alternatives match a specific set of criteria. This evaluator relies on alternatives from the input locator. It also learns from false alternatives to improve training.
An input locator provides a set of alternatives for each subfield. The Trainable Evaluator uses its training data to verify which of the input locator alternatives is the best result for the assigned subfield.
Input locators should provide input, but they should not be too complicated, as these will be executed during training and during extraction.
For example, your project contains a Format Locator that returns all of the amounts found on a document. You can use that Format Locator as an input to the Trainable Evaluator and train it to locate the "total" from the Format Locator amount alternatives.
The location of the Trainable Evaluator in the list of locators in a class is important. Only those locators listed above the Trainable Evaluator can be used as input locators. Since subfields are independent of each other, their order does not affect how they are processed. This means that you can organize the subfields in any way without affecting the output.
This evaluator is ideal when you want high out-of-the-box generic pre-trained extraction. Because you define input locators, you have a lot of influence over the alternatives that are considered for extraction. This is because you can configure and fine tune the input locators and script events. The Trainable Evaluator also learns quickly and with fewer samples, so you do not need a lot of training documents.
If however, you need specific extraction or fully automatic online learning, you should use the trainable group locators. This is because these locators allow you to deploy a project with zero training documents. The system learns from user changes and improves extraction over time.
Manage the Trainable Evaluator as follows:
The Trainable Evaluator window contains the following tabs: