Add to the mockup or plan based on visualization user stories about dianna HOT 7 CLOSED

See also

• Captum Insights and the
• LRP demos

for inspiration.

from dianna.

We created a mockup for a single view. The view we made shows a lot of information and could not run real time I think.

https://nlesc.sharepoint.com/:p:/s/DIANNA/EcPNpy-QJbdGgPSOVYgNoCIBZmea0tRHDswIXAh-qvQ43A?e=947awS

We are missing a lot of elements still.

from dianna.

The mock-up is presenting part of the "main UI" ideas. We think that an additional one allowing e.g. continuous "morphing" between triangles and circles and the corresponding XAI heat-maps for the model trained on our Triangles and Circles dataset could also be very useful, though with a secondary priority. The envisioned usage of this secondary UI targets 2 groups:

• XAI researches who want to study empirically the XAI methods' properties and
• our domain scientists, who (expected to posses healthy skepticism over XAI on top of their black-box model) want to get a feeling on how different XAI methods work on simple geometrical (circles and triangles) and a bit more realistic LeafSnap datasets before trying it on their (presumably complex) scientific data.

A predecessor of this UI is this simple old interface (back then for triangles and squares):

from dianna.

Another insinuation can come from the code for the "Benchmarking and Survey of Explanation Methods for Black Box Models" paper

from dianna.

Another inspiration can come from the code for the "Benchmarking and Survey of Explanation Methods for Black Box Models" paper

from dianna.

@elboyran and I have made a first proof of concept version of the mockup in https://github.com/dianna-ai/dianna/tree/41_dashboard.

Questions/Todo

1. How do we do colorbars? They may be different per explainer, and even per explainer-model combination. Do we even output colorbars in our current dianna.visualization module?
2. We need column/row labels.
3. We need a way to load models. File loader widget? Other?
4. We need a way to add data items
   1. For text: could just be a sentence, or what ever kind of text the model can take as input. So just a textbox.
   2. For images: need file loader widget again.
5. Colorbar should be colorbar/legend.
6. Class list needs to be added as well.

from dianna.

We should try out https://github.com/voila-dashboards/voila-gridstack for building the dashboard, looks quite convenient.

from dianna.