Quote:
Originally Posted by Jeff P
Personally, I like the idea of looking at numbers that reflect what I see happening in the race.
But that's a very different thing than the 'Is one more predictive than the other and does it have statistical significance?' question Steve is asking.
Imo, that (is it useful in a model) is what really matters.
-jp
.
|
I'm going for a long walk and will think about my response, so what follows may be nonsense.
It seems to me that a model with numeric variables would be more accurate than one with categorical variables? (At least in classical machine learning.) I understand we can 'covert' numbers to make them more amenable to the model (log, for example) but this can't be one of those cases.
If I fed the data to a neural network would it work with the actual numbers or convert them to abstractions that don't capture what's really happening? Hard to tell since it doesn't know what's really happening.
