Skip to content

Drifting Evaluation #4

@DavidSolan0

Description

@DavidSolan0

During the Drifting evaluation, I'll check two types of drift.

  • Data Drifting

Considering tabular data and only numeric features I'll propose a statistical hypothesis system approach to reveal distributional changes over time. It's likely that the multivariate distribution doesn't follow a standard form, so, I'll create a hypothesis system based on permutations and distances. The idea is to cover sudden, seasonal, and gradual changes.

For more complex solutions we can think to use VAE and its latent adjusted distribution to check distributional changes.

  • Concept Drifting

In the same vein. I'll create a pipeline that checks model performance (e.g., accuracy) and degradement during the time (e.g., chart control) that prevents the model to decay below established limits. It's important to remind the existence of curated data for this aim.

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions