Linear Regression: Machine Learning

Linear Regression
Machine Learning
Introduction
For this stage of development, we are trying to implement linear
regression and get to see it work on data.
We are performing this work on MATLAB for easy analysis of data and once
it is structured and tested, we will implement in Python for deployment.
Note: This implementation is just an experimentation for

qualitative analysis of Machine Learning algorithms and select the
best algorithm for the actual project.
Files included in this implementation

rev3.m - MATLAB script that steps we through the exercise
rev3data1.txt - Dataset for linear regression with one variable
plotData.m - Function to display the dataset
computeCost.m - Function to compute the cost of linear regression
gradientDescent.m - Function to run gradient descent
featureNormalize.m - Function to normalize features
1
For this programming part, we are only implementing the program to implement
linear regression with one variable.
1 Linear regression with one variable

In this part of this exercise, we will implement linear regression
with one variable to predict profits for a food truck. Suppose we are the
CEO of a restaurant franchise and are considering different cities for
opening a new outlet. The chain already has trucks in various cities and
we have data for profits and populations from the cities. We would like to
use this data to help we select which city to expand to next.
The file rev3data1.txt contains the dataset for our linear regression prob-
lem. The first column is the population of a city and the second column
is the profit of a food truck in that city. A negative value for profit
indicates a loss.
1.1 Plotting the Data

Before starting on any task, it is often useful to understand the data by
visualizing it. For this dataset, we can use a scatter plot to visualize the
data, since it has only two properties to plot (profit and population). (Many
other problems that we will encounter in real life are multi-dimensional and
can’t be plotted on a 2-d plot.)
In rev3.m, the dataset is loaded from the data file into the variables X
and y:
data = load('ex1data1.txt');%readcommaseparateddata
X = data(:, 1); y = data(:, 2);
m = length(y);%numberoftrainingexamples
Next, the script calls the plotData function to create a scatter plot of
the data. Wer job is to complete plotData.m to draw the plot; modify the
file and fill in the following code:
plot(x, y,'rx','MarkerSize', 10);%Plotthedata

ylabel('Profitin$10,000s');%Setthey −axislabel
xlabel('PopulationofCityin10,000s');%Setthex −axislabel
2
Now, when we continue to run rev3.m, our end result should look like
Figure 1, with the same red “x” markers and axis labels.
To learn more about the plot command, we can type help plot at the
MATLAB command prompt or to search online for plotting doc-
umentation. (To change the markers to red “x”, we used the option ‘rx’
together with the plot command, i.e., plot(..,[wer options here],..,
‘rx’); )
3
25
20
15
Profit in $10,000s
10
−5
4 6 8 10 12 14 16 18 20 22 24
Population of City in 10,000s
Figure 1: Scatter plot of training data
1.2 Gradient Descent

In this part, we will fit the linear regression parameters θ to our dataset
using gradient descent.
1.2.1 Update Equations

The objective of linear regression is to minimize the cost function
m
1 Σ. Σ2
J (θ) = h (xθ (i)) − y(i)
2m i=1
where the hypothesis hθ(x) is given by the linear model
hθ(x) = θT x = θ0 + θ1x1
Recall that the parameters of wer model are the θj values. These are
the values we will adjust to minimize cost J (θ). One way to do this is to
use the batch gradient descent algorithm. In batch gradient descent, each
iteration performs the update
4
1 Σ
m (i) (i) (i)
θj := θj − α (hθ (x ) − y )x j (simultaneously update θj for all j).
m i=1
With each step of gradient descent, wer parameters θj come closer to the
optimal values that will achieve the lowest cost J (θ).
Implementation Note: We store each example as a row in the the X

matrix in MATLAB. To take into account the intercept term (θ0), we add
an additional first column to X and set it to all ones. This allows us to
treat θ0 as simply another ‘feature’.
1.2.2 Implementation
In rev3.m, we have already set up the data for linear regression. In the
following lines, we add another dimension to our data to accommodate the
θ0 intercept term. We also initialize the initial parameters to 0 and the
learning rate alpha to 0.01.
X = [ones(m, 1), data(:,1)];%Addacolumnofonestox

theta = zeros(2, 1);%initializefittingparameters
iterations = 1500;
alpha = 0.01;
1.2.3 Computing the cost J (θ)

As we perform gradient descent to learn minimize the cost function J (θ),
it is helpful to monitor the convergence by computing the cost. In this
section, we will implement a function to calculate J (θ) so we can check the
convergence of wer gradient descent implementation.
The next task is to complete the code in the file computeCost.m, which
is a function that computes J (θ). As we are doing this, remember that the
variables X and y are not scalar values, but matrices whose rows represent
the examples from the training set.
Once we have completed the function, the next step in rev3.m will run
computeCost once using θ initialized to zeros, and we will see the cost
printed to the screen.
5
1.2.4 Gradient descent
Next, we will implement gradient descent in the file gradientDescent.m.
The loop structure has been written for we, and we only need to supply
the updates to θ within each iteration.
As we program, make sure we understand what we are trying to opti-
mize and what is being updated. Keep in mind that the cost J (θ) is parame-
terized by the vector θ, not X and y. That is, we minimize the value of J (θ)
by changing the values of the vector θ, not by changing X or y. Refer to the
equations in this handout and to the video lectures if we are uncertain.
A good way to verify that gradient descent is working correctly is to look
at the value of J (θ) and check that it is decreasing with each step. The
starter code for gradientDescent.m calls computeCost on every iteration
and prints the cost. Assuming we have implemented gradient descent and
computeCost correctly, wer value of J (θ) should never increase, and should
converge to a steady value by the end of the algorithm.
After we are finished, rev3.m will use our final parameters to plot the
linear fit. The result should look something like Figure 2:
Our final values for θ will also be used to make predictions on profits in
areas of 35,000 and 70,000 people. Note the way that the following lines in
rev3.m uses matrix multiplication, rather than explicit summation or
loop- ing, to calculate the predictions. This is an example of code
vectorization in Octave/MATLAB.
We should now submit wer solutions.
predict1 = [1, 3.5] * theta;

predict2 = [1, 7] * theta;
2.1 Debugging
Here are some things to keep in mind as we implement gradient descent:
• Octave/MATLAB array indices start from one, not zero. If we’re stor-
ing θ0 and θ1 in a vector called theta, the values will be theta(1) and
theta(2).
• If we are seeing many errors at runtime, inspect wer matrix operations
to make sure that we’re adding and multiplying matrices of compat-
ible dimensions. Printing the dimensions of variables with the size
command will help we debug.
6
25
20
15
Profit in $10,000s
10
Training data
0
Linear regression
−5
4 6 8 10 12 14 16 18 20 22 24
Population of City in 10,000s
Figure 2: Training data with linear regression fit
• By default, MATLAB interprets math operators to be matrix

operators. This is a common source of size incompatibility errors. If we
don’t want matrix multiplication, we need to add the “dot” notation
to specify this to MATLAB. For example, A*B does a matrix multiply,
while A.*B does an element-wise multiplication.
2.2 Visualizing J (θ)

To understand the cost function J (θ) better, we will now plot the cost over
a 2-dimensional grid of θ0 and θ1 values.
In the next step of rev3.m, there is code set up to calculate J (θ) over a
grid of values using the computeCost function that we wrote.
7
%initializeJ valstoamatrixof0's
J vals = zeros(length(theta0 vals), length(theta1 vals));
%FilloutJ vals
fori = 1:length(theta0 vals)
forj = 1:length(theta1 vals)
t = [theta0 vals(i); theta1 vals(j)];
J vals(i,j) = computeCost(x, y, t);
end
end
After these lines are executed, we will have a 2-D array of J (θ) values.
The script rev3.m will then use these values to produce surface and
contour plots of J (θ) using the surf and contour commands. The plots
should look something like Figure 3:
3.5
800
3
700
600 2.5
500 2
400
1.5
1

300
200 1
100
0.5
0
4
0
3 10
2 5 −0.5
1 0
0 −5 −1
−1 −10 −10 −8 −6 −4 −2 0 2 4 6 8 10
 0
1 0
(a) Surface (b) Contour, showing minimum
Figure 3: Cost function J (θ)
The purpose of these graphs is to show that how J (θ) varies with
changes in θ0 and θ1. The cost function J (θ) is bowl-shaped and has a global
mininum. (This is easier to see in the contour plot than in the 3D surface
plot). This minimum is the optimal point for θ0 and θ1, and each step of
gradient descent moves closer to this point.
8
9

Linear Regression: Machine Learning

Uploaded by

Copyright:

Available Formats

Linear Regression: Machine Learning

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Linear Regression: Machine Learning

Uploaded by

Copyright:

Available Formats

Linear Regression

Note: This implementation is just an experimentation for

Files included in this implementation

1 Linear regression with one variable

1.1 Plotting the Data

plot(x, y,'rx','MarkerSize', 10);%Plotthedata

Figure 1: Scatter plot of training data

1.2 Gradient Descent

1.2.1 Update Equations

where the hypothesis hθ(x) is given by the linear model

Implementation Note: We store each example as a row in the the X

X = [ones(m, 1), data(:,1)];%Addacolumnofonestox

1.2.3 Computing the cost J (θ)

We should now submit wer solutions.

predict1 = [1, 3.5] * theta;

Figure 2: Training data with linear regression fit

• By default, MATLAB interprets math operators to be matrix

2.2 Visualizing J (θ)

(a) Surface (b) Contour, showing minimum

Figure 3: Cost function J (θ)

You might also like