1. Generate the data set D as follows:
a. π·π· = 100
b. ππ = 25
c. ππ contains samples from a uniform distribution U(0,1).
d. π‘π‘ = sin(2ππππ) + ππ, where ππ contains samples from a Gaussian distribution N(0, ππ =0.3).
2. Select a set of permissible values for the regularization parameter ππ.
3. For each value of ππ, use the method of “linear regression with non-linear models” to fit Gaussian basis functions to each of the datasets. Use π π = 0.1.
4. Produce the plot as shown below, where
π·π·
1
ππ(Μ
π₯π₯) = ππ(ππ)(π₯π₯)
π·π·
ππ=1
ππ
1
(πππππππ π )2 = πππ₯π₯Μ
(ππ)− βπ₯π₯(ππ)2
ππ
ππ=1
ππ π·π·
1 1
π£π£πππ£π£πππππ£π£π£π£π£π£ = ππ(ππ)π₯π₯(ππ)− πππ₯π₯Μ
(ππ)2
ππ π·π· ππ=1 ππ=1
5. The test error curve is the average error for a test data set of 1000 points.