Polish Evaluation Tutorial

[brandenchan]

Proposed changes:

• Since the evaluation tutorial has gotten quite long and complex, we should make some minor changes to improve readability
• Added link to explain new metrics
• Added note that integrated eval mode runs as well when isolated eval mode is engaged
• Removed the initialize_device_settings
• Show there is no discrepancy in retriever isolated eval

To Do

• Reader eval discrepancy still exists

[review-notebook-app]

Check out this pull request on 

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB

[julian-risch]

Looks good to me so far but I think the filtering of no_answers needs more explanation and I would prefer to move it to a different line (earlier).

[julian-risch]

Here it was actually on purpose to have the output in the tutorial so that users can see the format of the Evaluation Report. I think we should keep it for that reason.

[brandenchan]

Sure, I have included the output from just those cells which print the Evaluation Report

[julian-risch]

LGTM! 👍 I also did a quick test run on colab and it works without any problems. With a fast internet connection in the office it took two minutes to download cross-encoder/stsb-roberta-large but other users might need to wait much longer. We could replace it with cross-encoder/stsb-roberta-base and mention in a comment that the large model would give even better results. Your choice. :)