› Use Transformation Designer › Manage table detection › Table training documents

Table training documents

When documents are first added to the Table Extraction Set, they are initially excluded from training and the Exclude from Training icon is displayed. Once a table label is added to a document it is included in training and the Include for Training icon is displayed. This means that documents are excluded from training until at least one table label is added to the training document. You can manually include a document with no table labels, but this may cause issues with training.

When it comes to training documents and table detection, note the following facts:

  • Every table in an included document is used for training, even unlabeled tables

  • Unlabeled tables are automatically labeled as "unknown"

  • Tables in excluded documents are not used for training, even if they are labeled

  • All tables in excluded documents are automatically labeled as "unknown"

    • Even if such an unlabeled table matches labeled tables in included documents

  • Table classification does not work well when two different tables have the same table label of when two tables of the same kind are assigned different labels

  • It is best to label all relevant tables in an included document OR exclude that document from training, altogether.

    If a document has no labels and it is included, it negatively affects the training if it contains tables that are similar to labeled tables elsewhere.

When editing a table training document one or more table models are required. If a table model does not exist when you start labeling your tables, it is possible to create one on-demand when editing a table training document. Table models are managed in the Project Settings - Table tab.

Each table label points to a specific table model. However, a table within a document can have additional settings other than the table label that points to the table model. It is possible for a single document to have several table labels that use the same table model, and it is also possible that each of these table labels have different settings.

What next?

Related topics: