CSCI 5622, Sec 001 Professor Mozer
Machine Learning Spring 2001
1
Assignment 7
Assigned: Thursday Apr 19, 2001
Due: Friday May 4, 2001
For this assignment, you have two choices: you can create a mini-project of your own choosing, or you can
enter the classifier competition.
Mini-project
For the mini project, I want you to pick a tractable yet interesting supervised learning problem and con-
struct a solution using either the software you’ve developed during the semester or by writing software to
implement other ML algorithms. Your work should include the following steps, which you should describe in
a brief write up:
1. Obtain a data set
2. Divide into training and test data
3. Determine a representation of the input and the output that is suitable to the problem. Use any knowl-
edge you have of the domain to derive a representation in which relevant information about the domain
is encoded explicitly, and will thus be easier for the ML system to exploit.
4. Determine an ML technique appropriate for the problem, and specialize it if possible based on your
knowledge of the domain. By “specialize,” I mean to design some type of bias into the algorithm that is
an appropriate bias for the domain.
5. Perform model selection to determine the best ML model based on your training data.
6. Evaluate performance of the ML system on the test data.
Classifier competition
For the classifier competition, I will provide you with preprocessed speech data and your task is to classify
an utterance into a response class indexed from 1 to 10. The training data consists of 4163 samples, and
the test data consists of 5730 samples, with roughly equal numbers in each class. Your task is to develop
the best classifier based on the training data, and then to send me predictions for the test set (which is not
labeled). I will report to you the accuracy your classifier obtains on the test set. You can use the training
data however you like to obtain the best classifier. The steps you take should be similar to those for the
mini-project.
Each training example will look something like this:
train4163 10
023 045 070 084 078 061 047 006 008 030
118 087 085 079 085 072 068 108 112 070
053 034 038 028 049 039 053 055 060 055
055 084 074 060 079 085 097 092 075 108
079 115 113 112 112 117 114 090 087 110
063 114 115 122 118 118 092 054 037 033
021 034 050 057 055 042 033 031 036 048
039 028 013 005 012 023 041 043 035 035
068 097 107 109 107 099 097 080 082 083
44
The label “train4163” is an identifier for the pattern. The next number, 10, is the class label (1-10) for the
example. The data then consists of 9 rows of 10 numbers, followed by a single number. Each number is
between 0 and 127. The columns in this array correspond to time slices: columns on the left are the begin-
ning of an utterance and columns on the right are the end of the utterance. The rows in this array corre-
