Hello,
I have a question for the EXIII for the part of n-grams. In the last question, it is said all parameters "a" have a value of 0.5 but in the solution, it is marked with 0.05.
And I would like to know why we consider the whole base of 729 possibilities but not 18 possibilities? What I have done is I add 0.5*18=9 +18 to get the denominator when calculating for exemple p(ut). But in the solution is 18+(0.05*729)=54.45.
Thank you in advance,
> In the last question, it is said all parameters "a" have a value of 0.5
It's a typo. which had been corrected in the solution but had not in the handout. Sorry.
It's now fixed.
> why we consider the whole base of 729 possibilities but not 18 possibilities?
The prior is on the whole set of possibilities, not only on the set of observed ones. That's actually it's "raison d'être"; considering only the corpus wouldn't make much sense (the reason of smoothing is precisely to go beyond the (non-)observed items; to prevent over-fitting (of the unseen))
Thanks for the answer.It is clear for me!
Follow the pulses of EPFL on social networks
© 2023 EPFL, all rights reserved