Skip to main content
Version: Next

Label Data

Updated 2024.05.05

A reliable dataset is crucial for training trustworthy models, and the system provides relabeling tools to help users create accurate datasets. Users can review the data included in the dataset and reset any incorrectly tagged values to the desired values, ensuring that the model is trained on corrected data.


Relabel Data​

Dataset → Actions → Label Data

The Label Data function is provided for AI Solutions that solve classification problems. To use the Label Data function, the following conditions must be met:

  • Edges, Local Dataset Source (S3 Dataset not supported)
  • The Solution assigned to the Dataset must support the relabel function (currently only supported for tabular solutions).

When creating a Dataset from the inference data of an Edge, you can refer to the model's judgment information (Score) and see the data that the model judged uncertainly. By labeling uncertain data, you can create a more precise training Dataset to prepare a more accurate model.

To relabel data​

  1. Navigate to the Edge Conductor and log in.
  2. In the left navigation pane, choose Dataset.
  3. Select the dataset to relabel.
  4. Click 'Actions' and select 'Label Data' (if 'Label Data' is disabled, the solution mapped with this dataset does not support the relabel function).
  5. Check the list of label values, and if necessary, edit label information (description, color, etc.).
  6. Check the label value set for each data in the dataset and modify it if necessary.

Edit Label Values​

Users can modify the items in the label list. You can change the name or display color of labels for the current dataset, add new labels, or delete existing labels. If the label information is changed, any data with the modified label must be re-labeled. Users can change the label value for each data from the AI inference result to a new label or further modify the label value for previously re-labeled results.

For example, you might initially label data as {Normal, ColorDefect} but later add a new label called AlignmentDefect, resulting in {Normal, ColorDefect, AlignmentDefect}. If the labels are changed to {Normal, AlignmentDefect}, any data previously labeled as {ColorDefect} must be re-labeled with an allowed value. The system will provide a warning guide to the user when saving the dataset.

To edit label values​

  1. Navigate to the Edge Conductor and log in.
  2. In the left navigation pane, choose Dataset.
  3. Select the dataset to edit.
  4. Click 'Actions' and select 'Edit Labels'.
  5. Modify the label names, descriptions, or colors as needed.
  6. If new labels are added or existing labels are modified, ensure all relevant data entries are re-labeled accordingly.
  7. Save the changes to update the dataset with the new label values.