dianna-ai / dianna

Deep Insight And Neural Network Analysis
https://dianna.readthedocs.io
Apache License 2.0
44 stars 13 forks source link

783 simplify overview tutorial #792

Closed elboyran closed 3 weeks ago

elboyran commented 4 weeks ago

Make the overview tutorial better.

review-notebook-app[bot] commented 4 weeks ago

Check out this pull request on  ReviewNB

See visual diffs & provide feedback on Jupyter Notebooks.


Powered by ReviewNB

elboyran commented 4 weeks ago

The ONNX model for the binary MNIST used in the tutorial is not the one from Zenodo! Some TF variant that is not on Zenodo is used, Using the one in Zenodo gives an error. So, the notebook still depends on local models and some other locally stored data.

elboyran commented 3 weeks ago

@elboyran thanks, looks very good 👍 I was able to run the notebook. One minor comment: is there a reason why we don't plot the results of kernelshap, the last method in the notebook?

@SarahAlidoost thanks for reviewing! If you recall during the last stand-up (or maybe you were not there?) there was the opinion that I should skip the whole Explainers section as it was redundant. Still, I opted for keeping it and gradually "simplifying" it, which includes firstly elaborating on LIME (corresponding to a missing LIME for Binary MNIST tutorial), then less for the second explainer, and finally not showing the last visualization. The point of the section for me is:

I hope this PR achieves that.