HW 4: due Thursday, December 1 in class
Submit your answers in hardcopy; also submit your program source code for
problem 1 via provide using the syntax
% provide comp167 hw4 myfilename1.here myfilename2.here
Make sure to include a readme file with instructions how to compile (if ap-
plicable) and run your code.
1. The file homework3.txt contains 10 samples; some are from ovarian
cancer tumors and somre are from breast cancer tumors. The claim is that
these two types of tumor samples produce microarray expression vectors
that are sufficiently different that they can be distinguished by clustering
methods. In particular, implement hierarchical clustering using the ordinary
(Euclidean) distance metric for computing inter-sample distances, and the
average-linkage clustering method of computing inter-cluster distances on
this dataset and output a resulting partition of the 10 samples into 2 types.
Submit your program source via provide as described above. Submit the
resulting cluster tree your program produces AND the two clusters you will
output in hardcopy.
2. The file cell-line-training.txt contains 10 samples, this time labelled
correctly as to whether they are ovarian or breast cancer samples. The
file cell-line-validation.txt contains another correctly-labeled 5 additional
examples. Using the k-nearest neighbors classifier (again with ordinary Eu-
clidean distance), assume the cell-line-training samples are correctly labelled,
and pretend the cell-line-validation class labels are unknown.
a) Compute the percentage of the cell-line-validation samples correctly clas-
sified based on the training data for k= 1,3,5 and 7, and report this in a
table.
b) Which value(s) of kdo best? Choose a value of kthat does best, call it
Kopt.
