Skip to content
This repository was archived by the owner on Jun 18, 2023. It is now read-only.

Training script "labels" too generic & not utilized #64

@ceholden

Description

@ceholden

Function get_training_inputs returns labels. These values are meant to be identification values for the pixel in the training data so that groups of pixels sharing the same label value can be grouped together for things like K-fold cross-validation. Currently, however, there are two problems:

  1. labels is too specific. The training data classification label, is a label, so this should be more specific to what it actually means
    • Could instead be, for example, roi_id, roi_label, etc.
  2. Nothing becomes of these labels! The SpatialKFold generators both use labels so that individuals from the same label are not mixed in the training / testing splits, but the labels are generated using scipy.ndimage.label instead of giving the user the ability to specify these labels via their ROI image.

Metadata

Metadata

Assignees

No one assigned

    Labels

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions