Data and model versioning

When you are collaborating with your team to iterate on your machine learning experiments, it is critical to be able to version your data and reproduce the model results. Each Model Run keeps a single source of truth for data that goes into your machine learning model. You can easily track your data versions and reproduce the model results.

What is versioned?

When you create a Model Run, Labelbox will automatically create a snapshot of data (includes data rows, labels, and data splits) that cannot be changed after you upload model predictions. Even if the Labels or Data Rows are updated later on, the original snapshot version of the data will be preserved.

After you upload model predictions or finish training a model with model training integration, Labelbox will automatically compute and version the model metrics for you.

Type

Versioned

Data

Data rows, labels, and data split

Model

Model metrics and model configs (coming soon)

How can I retrieve the data snapshot?

Follow the same steps in Train a model from Labelbox data to export data using the app or using the SDK.


Did this page help you?