The natural error function for the logistic
•
To fit a logistic model
using maximum
likelihood, we need to
minimize the negative
log probability of the
correct answer
summed over the
training set.
if t =0
if t =1
error derivative on
training case n