Models

In Trains, models are associated with the experiments that create them. The Trains Web-App (UI) allows you view model details, modify the configuration and class enumeration of status Draft (editable) models, and select different models when you reproduce experiments.

The models table

The models table columns in their default order are below. Dynamically order the columns by dragging a column heading to a new position. By default, the DESCRIPTION column is hidden. To show it, click and then select the DESCRIPTION checkbox.

  • MODEL - Model name. Use this field to sort the list.
  • CREATED - Elapsed time since the model creation. Hover over the elapsed time to view the date and time.
  • FRAMEWORK - The model framework, such as Caffe, TensorFlow, PyTorch, Keras, Custom. Use this field to sort and filter the list.
  • PUBLISHED - The model status which can be Published (read-only) or Draft (editable). Use this field to sort and filter the list.
  • TAGS - User-defined tags added to models for grouping and organization.
  • USER - The user name. This field is neither sortable nor a filter.
  • TASK - The Task name which is the name for the internal Trains representation and Python experiment script Task of the experiment creating the model. This field is neither sortable nor a filter.
  • DESCRIPTION - The model description.

Selecting models

The Trains Web-App (UI) provides several ways to select models.

To select a model:

  • In the Home page:
    • In the RECENT PROJECTS area, click the project card of the project containing the model. Click the MODELS tab. In the models table, click the model. The model details pane slides open.
    • In RECENT EXPERIMENTS, click the experiment containing the model. The experiment details pane slides open. In the ARTIFACTS tab, Input Model or Output Model area, click the model name. The model details pane slides open.
  • In the Projects page, click the project card or All projects. Click the MODELS tab. In the model table, click the model. The model details pane slides open.

Customizing the models table

You can customize the models table. Changes are persistent (cached in the browser), and represented in the URL so that you can save customized settings in a browser bookmark and share it with teammates.

Customize any combination of the following:

  • Dynamic column ordering - Drag a column title to a different position.
  • Hide columns - Click > clear the checkboxes of columns to hide.
  • Filter columns - Type of experiment, experiment status (state), user
  • Sort columns - Metrics and hyperparameters, type of experiment, experiment name, start and last update elapsed time, and last iteration.

Viewing and modifying models

Draft status models are editable. Published status models are read-only. You can only modify a status Draft model.

General model information

Model configuration

In status Draft models, you can modify the configuration.

  • In the NETWORK tab > Hover over the MODEL CONFIGURATION area > EDIT > Edit the configuration > SAVE.

Class enumeration

In status Draft models, you can modify the class enumeration.

  • In the LABELS tab > Hover over the LABELS area > EDIT > Edit the configuration > SAVE.

Read-only models (Publishing)

When you want to prevent changes to a model, make it read-only by Publishing it.

If a model is Published, you cannot reset experiments that write to that model (for example, reset to rerun experiments), because a reset cannot delete a read-only model. You could archive the model, however, and then reset to rerun the experiment.

  • In the models table, right click the experiment > Click Publish. The status changes to Published.

Downloading models

You can download a model.

  1. Open the model context menu.
    • In the model table, right click.
    • In the model details pane, click (menu).
  2. Click Download.

Moving models to other projects

To better organize your work, you can move models between projects.

  1. Open the model context menu.
    • In the model table, right click.
    • In the model details pane, click (menu).
  2. Click Move To Project > Select a project > MOVE.

Archiving models

Keep the project page models table clear of the models you do not actively use by moving them to archive.

Archive models using one of the following methods:

  • In the models table, archive one model - Right click the model Archive > ARCHIVE.
  • In the models table, archive multiple models - Select the models checkboxes > Archive > ARCHIVE.
  • In the details pane, archive one model - Click (menu) > Archive > ARCHIVE.

Ignore the comparison footer

The comparison footer appears when you select more than one experiment. For archiving and restoring, ignore it.

To restore models:

Restore models from the archive using one of the following methods:

  1. Click OPEN ARCHIVE
  2. Restore one or multiple models:
    • In the models table, restore one model - Right click the experiment Restore > RESTORE.
    • In the models table, restore multiple models - Select the models checkboxes > Restore > RESTORE.
    • In the details pane, restore one model - Click (menu) > Restore > RESTORE.

Locating the creating experiment

You can open the experiment that created the model from the model details, GENERAL tab.

  • In the GENERAL tab, in CREATING EXPERIMENT, click the experiment name. The experiment table appears, showing the experiment's details pane.

Locating the model ID

  • In the details pane, top area, click ID. The model ID appears.