Closed rfl-urbaniak closed 7 months ago
Note the comment in central_park_birds_predictors.ipynb
:
# for instance, it seems that as a proximity parameter setting
# 19 should work well for sparrows, less so ducks
# 46 should work well for ducks, not terribly for sparrows
# 85 should not work very well for either
This justifies the choice of hyper-pamaters that I used and allows for a slightly less cluttered histogram. Will keep thinking about how to best visualize this comparison.
Leaving the hyperparam choices the way you had them, both here and in the other PR that contains the inference notebook - we can discuss and potentially revise later.
Leaving the hyperparam choices the way you had them, both here and in the other PR that contains the inference notebook - we can discuss and potentially revise later.
Thanks, yes, happy to discuss. To my eye, the numbers 19 and 46 don't stand out as particularly different between the histograms for the different groups of birds.
Looks good! I just see the predictors notebook, not the downstream inference one -- is the plan to include this later?
Yup, as suggested by PR titles. :)
There is a separate inference notebook, downstream. Some of the specific functions are still left in the cpb module, perhaps it might be worth refactoring and unifying them, but at this point these are the ones that don't play well with random hungry followers, so I'll just make an enhancement issue about this.