Skip to content

Segmenter class (exposing strided-roll creation) #15

Open

Description

@jvdd

Create a Segmenter class, this class exposes (in a clean interface) the creation of strolls (from the StridedRolling class).

Such a class enables users to create efficiently strided rolls (strolls), opening the door towards integration with;

  • deep learning packages (e.g., tsai, 1D or multi-dimensional convolutions)
  • raw time-series packages (e.g., tslearn, sktime)

Activity

  1. jvdd commented on Dec 22, 2021

    @jvdd
    CollaboratorAuthor

    I think this should go even further (near full decoupling of Segmentation and Feature Extraction functionality)

  2. jonasvdd commented on Jun 7, 2022

    @jonasvdd
    Collaborator

    Also:

    • add examples on how to use tsflex for already segmented data
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    enhancementNew feature or request

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions