Skip to content

Ideas for experiments - master's thesis #8

@she3r

Description

@she3r

We currently have two options:

  1. To what extent does the interactive layer help in generating counterfactuals? This option involves collaborating with other people to conduct research using the Streamlit application. We need to come up with a meaningful study that we can assign to students, one that can yield useful results, considering that students have no interest in obtaining good counterfactuals—it is only right if the experiment does not end on our testers clicking random values on the screen.
  2. Option without the participation of others - experiments for the ACFX API, where EBM is not the only blackbox available- Comparison for ACFX/ACFXLinear (ACFXCustom). I could potentially experiment using Friedman post-hoc statistical tests as you did, but this time comparing different input tools for acfx (classifier/optimizer). Similarly, an ablation study could be conducted, but for different acfx.
    Option 2 - that one I think I am ready to do without further brainstorming, but it is a worse option because the streamlit app has no use here.
    Option 1 - requires further consideration. Let's brainstorm below.

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