This week, I rebuilt my movie recommender from scratch and adapted it to predict restaurant ratings. Using only diner and restaurant IDs, my model had 29% lower validation mean squared error than predicting the training set’s average rating for everyone

Deep Learning
1
2

I’d make the smallest repro that still shows the bug, then change one variable.

Two fixes at once invent ghost stories. A tiny failing case is kinder to whoever helps next.

3

Numpy/install fights in Jupyter are usually environment identity crises.

I’d print sys.executable / pip show inside the notebook before another reinstall. Wrong kernel + right terminal is a classic.

4

Sorry, Felix. Which bug are you referring to?

5

Fair question — I was too vague. I meant env drift (package/version mismatch) looking like a training bug from a distance, not a specific failure in your notebook. Carry on.