Description
1 Overview
In this project, you will experiment with linear regression, overfitting, feature “engineering”, and basic matrix operations.
You are given a dataset, consisting of 926 examples. Each example has 8 real-valued
predictor attributes x (“regressors”), and a single real-valued dependent value y to be
predicted. Your goal is to use this data to build a function that will predict values y for
new data x. Specifically, we have more data that was generated from the same source. We
will measure how well your predictor does on this new data.
2 Data
You are given a file, traindata.txt. This file contains 926 rows, each with 9 numbers,
meaning it constitutes a 926×9 matrix. The first 8 columns contain the data that you will
be predicting from, while the last column contain the data that you will predict.
As an example, to read this data in matlab, you might do:
traindata = importdata(’traindata.txt’);
X = traindata(:,1:8);
Y = traindata(:,9);
We also provide a file testinputs.txt. This contains 103 rows, each with 8 numbers.
This is just like above, except that we do not provide the true output value. You might
read this in with something like:
Xtest = importdata(’testinputs.txt’);
3 Task
Your goal in this project is to provide a text file with 103 numbers in it, consisting of your
predictions for each of the 103 test inputs. Specifically, we will measure the mean-squared
error of your predictions,
1
Linear Feature Engineering 2
1
103
X
103
i=1
(yi − y
∗
i
)
2
where yi
is your prediction, and y
∗
i
is the true output value.
4 Methods
To make these predictions, you should create features. That is, given and input x (of
length 8), you should make a function which creates a (presumably longer) vector x
0
.
Then, you will fit a linear regression model of y to x
0
.
5 Deliverables
1. Code. Please submit all code for this project, along with a very brief README.txt
which explains to us how to use it. We should be able to run your code and regenerate your predicted values.
2. Report. Two page maximum, single column. Please be clear and concise. You
should discuss the following topics:
(a) At the top of your report, give two numbers:
i. The training error 1
926
P926
i=1(yi − y
∗
i
)
2
, where yi
is the predicted value, and
y
∗
i
is the data value.
ii. Your prediction for the test error 1
103
P103
i=1(yi − y
∗
i
)
2
.
(b) How did you choose what features to use?
(c) How did you arrive at your prediction for the test error?
(d) How did you deal with overfitting?
3. Presentation. Give a 8-minute presentation, describing your experiences, what you
tried, what worked, what didn’t work, what features you use, and how well you
estimate your team did in terms of least-squares error.
6 Grading:
• Content: 50%
• Report: 20%
• Presentation 30% (including Q&A)
