- turn in hw2 on paper on Monday (no need to use handin)
- hw2 has been modified. The first question is now about Bayes rule.
If it looks like Q1 is about linear regression you have hte old version.
Please be sure you print out the up to date version!
- Q2. The proof can consist of pictures, but it should be convincing,
not handwavy.
- Q3 part 1.
Train on Xtrain(1:10,:) and evalyate mse on Xtrain(1:10, :) and on
Xtest(:,:).
Then train on Xtrain(1:20,:) and evalyate mse on Xtrain(1:20, :) and
on Xtest(:,:).
etc. Plot mse vs training set size.
- Q3
[w, mu, sigma] = train_ls(...)
here w is a vector of [w0 w1 ... w8] in the equation on p1
- Q5 part 2.
The question should read "*In general*, does cross validation with
small k tend to choose models that are too simple or too complex?"
This tendency may not be exhibited by this particular daataset/ model,
since it is quite small. "Small k" means k=2 or 3. "Large k" means k
is, say, 10% of N (the size of the data), so if N=1000 (say), large k
would be say k=100.