Lets say that we are developing a gradient ascent/descent algorithm to learn the GLM with Poisson error from the last question. All of the above answers (except for none of the above) had the term

Why can we drop this term and not consider it when developing our algorithm?

A. Because for large yi, this term is so tiny it does not affect the answer.

B. Because this term does not depend on the regression coefficients, which is what we are maximizing over.

C. Because gradient ascent/descent cannot work over two variables at the same time.

D. Because there is no way to simultaneously maximize over all of the regression coefficients and yi at the same time.

E. None of the above.

logy!1