How to Start a New Label Generation

Label generation is the first step in equipping your human reviewers with super-human speed and scale. We recommend generating labels iteratively as described here.


To generate labels

  1. Start a New Label Generation
    • Go to the dashboard.
    • Click "Label Generation".
    • Click on "Set Up New Generation."
  2. Choose the project
    • This defaults to the current project you're on.
  3. Choose an Overwrite setting
    • Overwrite Generated Labels. This is the default. It will regenerate any labels your team hasn't reviewed yet, and leave any human-reviewed results as-is.
    • Overwrite Generated and Human Labels. This setting will regenerate all labels, even the ones your team has reviewed.
  4. Task
    • "All" is the default and usually the correct choice. You can choose to run only part of the label generation process, which can be helpful for debugging.
  5. Generation Size and Parameters.
    1. Batch of patients
      1. This selection will generate labels for a random "Batch" of patients that hasn't been generated yet.
      2. Choose the Batch size. We recommend starting with 1 to verify results, and increasing from there.
      3. You can also choose to re-run generation for an existing batch, or clear batches from your dataset.
    2. Single specific patient.
      1. If you are focused on a single patient, you can choose this option, and enter the patient ID to generate for that patient.
      2. In this option, you can choose to only generate a specific variable or a specific document if you're debugging.
    3. All patients
      1. This will generate for all patients.
      2. In this option, you can choose to only generate a specific variable if you're debugging.
      3. NOTE: Generating for All Patients can be data- and time-intensive for large datasets. We recommend working in batches when possible.
  6. Click on the "Generate" button at the bottom to generate labels.
Did this answer your question? Thanks for the feedback There was a problem submitting your feedback. Please try again later.