Understand Precision Recall Curve

Get an overview of precision recall tradeoff on varying confidence thresholds

Description

With Nucleus you can obtain a precision recall curve to understand how model performance varies when the confidence thresholds are tuned. This can enable you to have a more nuanced understanding of model performance than by using a single metric.

Pre-reqs: Dataset, Annotations, Predictions (evaluated)

Steps

  1. Open dataset with both annotations & predictions uploaded
  2. Use the top navigation to go to the “Charts” page
  3. Use the drop down to select model of interest
  4. Scroll down to “PR Curve” chart
  5. Hover on any point to get a precision and recall value pair
  6. Limit to a specific class by selecting class from the chart drop down
  7. Click on any plot point to query the underlying data items in the grid view