Update proponent/opponents visualization functions

dilyabareeva / quanda

A toolkit for quantitative evaluation of data attribution methods.

https://quanda.readthedocs.io

MIT License

33 stars 0 forks source link

Closed dilyabareeva closed 2 months ago

dilyabareeva commented 2 months ago

currently in tutorials/utils/visualization.py
add predicted labels to the samples
make margins and spacing even
increase font
no enumeration for proponent/opponent 1-3
maybe take inspiration from https://github.com/MadryLab/trak/blob/main/assets/imagenet_figure.jpeg (but not do a total rip-off)

dilyabareeva commented 2 months ago

The relevant tutorial is now in tutorials/demo.ipynb

dilyabareeva commented 2 months ago

The demo.ipynb was renamed to demo_explainers.ipynb. The target test indices are loaded to test_indices variable