Labels as Features#

In the platform, input features and labels are treated with the same level of consistency and rigor, reflecting their analogous roles in the machine learning workflow. Both are computed per sample and shuffled together during the training process, ensuring that the integrity and alignment between features and labels are preserved.

This unified treatment eliminates the need for special handling or separate processing steps for labels, simplifying the overall workflow and reducing potential sources of error. By treating labels as seamlessly integrated with features, the platform ensures that the model learns from a cohesive dataset where input-output relationships are maintained accurately.

This approach enhances the model’s ability to learn meaningful patterns and relationships, leading to more robust and reliable predictions. Additionally, it streamlines the development process, making it easier for users to manage and iterate on their models without being bogged down by disparate handling of features and labels.

Label visualization#

Because labels are treated as features, the same powerful visualization capabilities are available to both. By applying the same visualization tools used for input features to labels, users can gain deep insights into the data's underlying patterns and relationships.

UI Hints#

The programmer/author of the label has rich toolset for providing annotation hints to the chart. The platform will dynamically check for a get_ui_hints function in the feature. If it finds one, it runs it, and displays the annotations as specified by the output of that function.

Mean Label#

Inconspicously placed just below the chart is the powerful Mean Label control. When the button is clicked, the platform will take random samples from the selected raw data, compute the labels from these samples, and display the mean value.

Knowing the mean value of a label in a training dataset is valuable because it provides a baseline for understanding the overall trend or central tendency of the data, which can help in evaluating the model's performance and identifying potential biases or anomalies.