[timhndrxn]

Exercise 3.2 does not seem clear concerning the test set of size 1,000 mentioned at the end. The sentence begins "Similarly" implying that this test set becomes a data set D of size N = 1,000 to be made not linearly separable. Then you repeat the whole experiment on the new data set D.

But, because the purpose is to examing E_out, perhaps the intent is to take 1,000 points that are (very likely to be) out of sample, and evaluate w-vector vs. w-hat-vector on them.