Skip to content

Manage models

Overview

Models are the text analysis engines produced by either running experiments or importing CPK files.

To access the list of all the models generated for the project, select the Models tab.

The tab is divided in the following panels:

  • The model list on the left side with the related summarized information, such as:
    • Model name
    • Execution time
    • Training library name
    • Documents number
    • Annotated documents number
    • Publishing status
  • The model details and the knowledge graph navigation regarding the selected model in the middle.
  • Possible operations with the selected model and the related experiments on the right side.

Change the model in focus

To change the model in focus, select it in the list on the left.

Sort the model list based on an item

To sort the model list based on an item, select the related dropdown menu at the top of the list then the criteria of your interest.

Filter the list based on an engine

To filter the list based on an engine select the related dropdown menu at the top of the list, then the in the Engine area select the engine of your interest.

Filter the list based on a training library

To filter the list based on an training library select the related dropdown menu at the top of the list, then the in the Train library area select the library of your interest.

Search for a model

To search for a model enter the search criteria (minimum three characters) in the search bar then press enter.

Select the X icon to reset the search criteria.

Upload a CPK

To upload a CPK file, select Upload CPK .

Once a CPK is uploaded the remapping panel is automatically displayed.

Remapping

After importing the CPK, the remapping panel is focused on the Remapped tab and it displays the result of the automatic remapping that it could be also empty in the worst case.

Since the imported CPK model may differ from the project you are working on, for example because it is produced at different times or by different users, it is sometimes necessary to remap it.

To perform the project remapping use the following tabs.

Info

It is also possible remap the resource later selecting Remap model.

All resources

This tab shows the complete resources of the imported model and if and how they have been remapped.

If no automatic remapping occurred, no match is displayed. Therefore, it is be possible to perform a manual remapping, if it is considered appropriate.

If you want to manually remap a resource from the imported model to a category of the existing project:

  1. Select a class from the list box on the right side.
  2. Select Save to confirm.

This mapping will be displayed in the Remapped tab.

Remapped

This tab shows the remapped items, both manually and automatically.

If you want to edit the remapping follow the same procedure described in All resources.

To quit the remapping panel select Exit.

Missed

This tab shows the imported model resources that don't have a mapping with the existing model resources.

If you need to remap a resource from the imported model to a category of the existing project:

  1. Select a class from the list box on the right side.
  2. Select Save to confirm.

This mapping will be displayed in the Remapped tab.

To quit the remapping panel select Exit.

Not Mapped

This tab shows the existing resources that don't have any mapping with the imported resources. To quit the remapping panel select Exit.

The model details

Select the Model details tab (displayed by default when you access the Models tab) to display information regarding:

  • The Training library (in the upper part).
  • The Training settings. Select the Train settings tab.
  • The AutoML values, in case of AutoML parameters setting. Select the AutoML values tab.
  • The resources, in case of Auto-ML extraction experiments. Select the Resources tab.

If the model in focus in an uploaded CPK the panel shows information about Knowledge graph navigation

Select the Knowledge graph tab to navigate the knowledge graph used in the model in focus.

Note

Only the list view is available.

Start an experiment with the selected model

To start an experiment with the selected model in focus:

  1. Select a model or a CPK from the model list.
  2. Select Quick start an experiment.
  3. In the dialog, optionally type a name, select the library, then select Next.
  4. If the experiment you are starting:

    • Is not the first one of your project.
    • Is performed with the same model type of one of your previous experiments.

    The Select run preset dialog appears.

    In this dialog:

    1. Select:

      • Last experiment run settings to start an experiment with the same settings, libraries and engine of the previous experiment with the same model type you selected from the model list.

      Or:

      • Default run settings to start a new experiment.
    2. Select Edit settings or Fast start to go to the next step.

  5. Select the matching strategy.

  6. If your project has sections and you selected a model made with the Explainable Extraction type or a CPK, select:

    • Discard document: if there are documents with sections that are not compatible with the model in focus, the document will be discarded.

    Or:

    • Use default section: if there are documents with sections that are not compatible with the model in focus, such sections will not be considered in the experiment.

    If your project has documents with layout and you selected a model made with the Explainable Extraction type or a CPK, select:

    • Require layout information to only analyze documents with layout information.

    Or:

    • Use layout when available to analyze documents with layout information when available.

    Or:

    • Plain text to analyze documents without considering layout information.

    Note

    If your model has both documents with layout and sections, in the Experiment settings, Summary area, these options are available respectively in the Layout and Sections sub-tabs. The Matching strategy sub-tab is also available to select the matching strategy.

  7. Select Start.

Start an experiment with no models

To start an experiment with no models in the Models tab, select Start an experiment, then follow the procedure described in the dedicated page.

Create a knowledge graph customization project

To create a knowledge graph customization project:

  1. Select a CPK or a model produced with explainable extraction from the model list.
  2. Select knowledge graph Customization.
  3. In the dialog, type a name, an optional description and select Create.

Note

The project creation will occur as a background task.

Note

The knowledge graph customization project on a CPK can be created only with compatible tech versions.

Publish and unpublish a model

To publish a model, select it from the models list and, if not a CPK, then select Publish model.

Once a model or a CPK is published, it is marked with a green dot in the models list.

Published models are available to be used in the workflows of the NL Flow part.

To unpublish a model, select Unpublish model on the right panel.

Remap a model (for CPKs only)

Select Remap model if you need to remap the resources (classes in the extractions projects) between the imported CPK and the existing project.

Export a model

To export a model:

  1. Select Export model if the model is generated with an Auto-ML Extraction engine or Export CPK if uploaded.
  2. In the Export model window (or Export CPK window), type the model name in File name or confirm the suggested one, then select:

    • Export model

    Or:

    • Export CPK (in case of a CPK)
  3. Select Download either in the dialog or in the notification in the lower right corner.

Edit the model name

To edit a build name:

  1. Select Edit model name and the Edit model name dialog is displayed.
  2. Enter the new name in Model name and select Save to confirm.

Delete a model

To delete a model:

  1. Select Delete model and the Delete model dialog is displayed.
  2. Select Delete to confirm.

To check related experiments, if any, see the Related experiments sub-panel.

Double-click the experiment to view its quality.