How can a newbie retrieve, as a prediction of a binary (sale/no sale) variable, a representation of 0-10 based on the likeliness of the new record being a sale? For example, given this record of customer data, I identify the likelihood of a sale as an "8.5". I can do basic pred, just don't how how to get it as a score.
#Prediction Score 0-10?
deft dawn
random hill
If you are doing binary (yes/no) classification, you'd usually have a softmax layer at the very end. So if you specify sale as 1, and no-sale as 0, the network will actually return to you a float between 0.0-1.0.
With some care, that float can be treated as a probability, or confidence the network has in the prediction, so you can interpret 0.85 as "8.5", or "pretty confident".
deft dawn
And this is somehow superior to rf.predict_proba(X_test) ?
random hill
well, you are talking about an implementation, and I'm talking about the theory 🙂
if you are using predict_proba, how about you multiply it by 10 and call it a "score" then?
deft dawn
Likely will, thanks for the help 👍