search

carpentries-incubator / deep-learning-intro

Learn Deep Learning with Python
https://carpentries-incubator.github.io/deep-learning-intro/
Other
30 stars 36 forks source link

Improve solution to "The Training Curve" exercise #278

Closed tobyhodges closed 11 months ago

tobyhodges commented 1 year ago

(With apologies if there is an issue already open about this.)

In my view, the solution provided for the The Training Curve exercise could be improved. At the moment, the exercise and the text before it talk about the potential value of plotting out the loss curve, as a way to evaluate the quality of the training configuration. However, I think it leaves unanswered the question many learners would have at this point - "what does a poorly-configured training process look like and how can I fix it?" I think it could be improved by adding an example of a loss plot for a poorly-configured training process, and by explaining what "jitter" in the curve might indicate/how that could be addressed, etc. Perhaps you could even set another exercise that requires them to generate a loss curve for a poorly-configured training process? (Or that might take too long to achieve?)

dsmits commented 1 year ago

Thanks @tobyhodges, you make a good point.