Thank you!
I was wondering about question 8. It was asked about identifying the incorrect statement, however, the only correct statement is marked to be the correct answer?
the box is ticked as it should. sorry for the 4 confusing answers labels, they refer to correctness of the box, not the statement. but all seems ok to me
Could you please give more details on the complexity of the K cross validation ?
We are training a dataset on K times on the union of K-1 groups of N/K samples. It's not very clear whether the complexity should be O(K) or O(K(K-1)).
Take N=K, we train on K-1 samples and test on 1 sample. We do it K times. Here we obviously have O(K(K-1))
Could you please give more details on the complexity of the K cross validation ?
We are training a dataset on K times on the union of K-1 groups of N/K samples. It's not very clear whether the complexity should be O(K) or O(K(K-1)).
Take N=K, we train on K-1 samples and test on 1 sample. We do it K times. Here we obviously have O(K(K-1))
I think there is the description that "the training samples are sufficient large". So you cannot really make the assumption that N=K because this is not possible asymptotically.
@Anonymous said:
Hi, could anyone explain a bit about in question 29, why the eigenvalues are sorted in a non-increasing order, instead of a non-decreasing order?
Non-decreasing means \(\geq\), while non-increasing means \(\leq\)
@Anonymous said:
Hi, could anyone explain a bit about in question 29, why the eigenvalues are sorted in a non-increasing order, instead of a non-decreasing order?
Non-decreasing means \(\geq\), while non-increasing means \(\leq\)
@Anonymous said:
Hi, could anyone explain a bit about in question 29, why the eigenvalues are sorted in a non-increasing order, instead of a non-decreasing order?
Non-decreasing means \(\geq\), while non-increasing means \(\leq\)
Yeah, but shouldn't the largest eigenvalues come first? like 5, 4, 3, 2 ,1 --- non-decreasing order
Exam and solutions of today are available on github
and on the course website
4
Thank you!
I was wondering about question 8. It was asked about identifying the incorrect statement, however, the only correct statement is marked to be the correct answer?
the box is ticked as it should. sorry for the 4 confusing answers labels, they refer to correctness of the box, not the statement. but all seems ok to me
Hi,
Could you please give more details on the complexity of the K cross validation ?
We are training a dataset on K times on the union of K-1 groups of N/K samples. It's not very clear whether the complexity should be O(K) or O(K(K-1)).
Take N=K, we train on K-1 samples and test on 1 sample. We do it K times. Here we obviously have O(K(K-1))
1
I think there is the description that "the training samples are sufficient large". So you cannot really make the assumption that N=K because this is not possible asymptotically.
Yes but in the end you still train a union of (K-1)*N/K elements for each group.
In the correction it says we train (K-1) / K elements but it would mean less than 1 ?
Thanks for the amazing work publishing this so early, really appreciate it! You've calmed my nerves massively! <3
3
Thank you so much for this fast respond! I truly appreciate it
Hi, could anyone explain a bit about in question 29, why the eigenvalues are sorted in a non-increasing order, instead of a non-decreasing order?
Non-decreasing means \(\geq\), while non-increasing means \(\leq\)
other way around
Yeah, but shouldn't the largest eigenvalues come first? like 5, 4, 3, 2 ,1 --- non-decreasing order
5, 4, 3, 2, 1 is non-increasing order: 5 ≥ 4 ≥ 3...
In the first part of the last open question, you prove that B <= A, and in the previous question, we show that t<= A. It does not imply that t <= B.
Ah, I get it now, thanks!
Add comment