Create a model

Developer guide: Create a model via the Python SDK

A model experiment is a directory where you create, manage, and compare a set of model experiment runs related to a machine learning task. A machine learning task is specified by the ontology of data that you use for integration with a model training service.

Labelbox products share a common structure of data rows, labels, and features. For example, in the diagram below, the global schema definition of a feature is used in ontologies for labeling data as well as model training, testing, and evaluation in the Model product. Labelbox is structured so you can reuse feature schemas to create ontologies specific for individual data labeling or model training tasks.

Create a model experiment

📘

This section of documentation is in progress.

We have built a new way to build model experiments in Labelbox UI that allows you to send data and annotations to model runs directly from Catalog. We will update the documentation soon.

To create a new model experiment:

1. Go to Model , Experiment and click Create then select Model experiment.
2. Configure the model experiment by giving it a name and a thumbnail.
3. Select an ontology.
   The ontology determines which data rows can be used for model training. Once you select an ontology, all data rows containing the annotations within the selected ontology in your entire Catalog will be available for inclusion. You can further narrow the scope by utilizing projects or datasets filters.
4. Click Create model with (n) data rows to create the model.

Delete model experiment

To delete a model experiment:

1. Use Model to open the model gallery and then select Experiments to display available model experiments.

2. Select the model experiment you want to delete and then open the Settings tab.

3. Select Danger zone and then select the Delete model button.

4. When the Delete model confirmation prompt appears, select Delete.

Note that model experiment deletion is permanent and cannot be undone.