​

What is a model run?

A model run represents a single iteration within a model training experiment. Each model run contains a versioned snapshot of the data rows, annotations (predictions and/or ground truth), and data splits for each iteration within a model training experiment. Model runs make it easy for you to reproduce a model training experiment using different parameters. You can also use model runs to track and compare model runs trained on different data versions. Each model run contains the following configurations: Note: When you configure the parameters above in a model run, you have the option to save the configuration as a model run config. Model run configs can easily be reused for future model runs.

​

Create a model run

To learn how to create a model run in an experiment, see Create a model run.

​

View/manage model runs

To view and manage the model runs in an experiment, go to Model and select an experiment. The Model runs subtab is the default view when you select an experiment. Within the model runs tab, you’ll see the following tools to help you navigate and narrow down the data rows in your model run.

​

Filter and sort

When you compare two model runs, you can use the filters to further narrow down the data rows in the model runs being compared. Then, you can save these filters as a slice to reuse later. Customers often use filters to find low-performing slices of data, surface labeling mistakes, and identify high-impact data to use for relabeling. For details on filtering and sorting data rows, see the following pages: For computer vision-related data rows, you can also use the confidence and IOU threshold settings to modify your view of the data rows (gallery view and metrics view only).

​

Gallery view

You can use the gallery view to:

1. Visually compare predictions against ground truth within a single model run.
2. Visually compare a model run’s predictions and/or ground truth labels against another model run.

​

Display settings

You can customize how the data rows are displayed in the gallery view by opening the Display panel (the eye icon). From there, you can customize the following settings:

​

List view

Use the list view to view each data row asset, predictions, and annotations information in each row.

​

Metrics view

Use the metrics view to analyze the distribution of annotations and predictions in your data, evaluate the performance of a model, and quantitatively compare two models. See Model metrics for details on which metrics are provided and how to use the metrics view.

​

Projector view

The projector view uses embeddings to display your data rows in a 2D space. This view allows you to select clusters of data points in order to uncover patterns and diagnose systemic model and labeling errors. To switch to the projector view, select the cluster icon in the corner. You can use the projector tool to employ dimensionality reduction algorithms to explore embeddings in 2D interactively. The projector view supports two algorithms for dimensionality reduction: UMAP & PCA.

​

Update/delete a model run

Click the settings icon in the top right corner of the model runs page to modify the selected model run. If you choose to delete the model run, you will be prompted to confirm the deletion. You can also update a model run by updating the Model run config file. To learn how to update the Model run config file, read our docs on Model run config.