Scribd
Upload
20 views5 pages

2023 Spring ML HW01

Uploaded by

Sushiman
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
20 views5 pages

2023 Spring ML HW01

Uploaded by

Sushiman
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 5

ML Homework1

Description :
Please write a program that can do regularized linear model regression (polynomial basis) and
visualization.

You should do it by closed-form LSE approach, Steepest descent, and Newton's method.

Input parameters:

1. the path and name of a file which consists of data points (comma seperated: x,y):

1,12
122,34
-12,323
...

2. the number of polynomial bases n.

3. lambda (only for LSE case)


Program Behavior: For example, if the number of bases is set to 3, it means that the program
is going to find a curve that best fits the data points by
Required functions:

a. For closed-form LSE approach:


1. Use LU decomposition to find the inverse of , Gauss-Jordan
elimination will also be accepted.(A is the design matrix).
2. Print out the equation of the best fitting line and the error.
b. For steepest descent method:
1. Use steepest descent with LSE and L1 norm to find the best fitting line.
2. Print out the equation of the best fitting line and the error.
3. reference (hint : Consider using a smaller learning rate.)
c. For Newton's method:

1. Please use the method mentioned in the lesson.


2. Print out the equation of the best fitting line and the error, and compare to LSE.
d. For visualization:
1. Please visualize the data points which are the input of program, and the best
fitting curve.
2. It's free to use any existing package.
NOTE:

Use whatever programming language you prefer.


You should use as few functions from any library as possible. That would be great if you
implement all detail operations (like matrix operations) by yourself.
Time complexity is not what we care for now, but if you like to improve it in that regard,
it is always good for you.

Grading policy

You must implement matrix inverse operation by yourself. Please do not use the built in
package or you'll not get 100.

Sample input & output (for reference only, please note that it doesn’t include results from the
steepest descent method)
Input: A file (here shows the content of the file)

$ cat testfile.txt
-5.0,51.76405234596766
-4.795918367346939,45.42306433039972
-4.591836734693878,41.274448104888755
-3.979591836734694,26.636216497466364
-3.571428571428571,20.256806057008426
-2.9591836734693877,11.618429243797276
-2.7551020408163263,10.450525068812203
-1.7346938775510203,1.8480982318414874
-1.3265306122448979,-1.0405349639051173
-0.9183673469387754,-4.614630798757861
-0.7142857142857144,-1.3871977310902517
-0.3061224489795915,-1.9916444039966117
0.1020408163265305,-0.912924608376358
0.7142857142857144,6.63482003068499
1.1224489795918373,9.546867459016372
1.7346938775510203,15.72016146597016
1.9387755102040813,20.62251683859554
2.5510204081632653,33.48059725819715
2.959183673469388,40.76391965675495
3.979591836734695,66.8997605629381
4.387755102040817,78.44316465660981
4.591836734693878,86.99156782355371
5.0,99.78725971978604

Output
Case 1: n = 2, =0

LSE:
Fitting line: 4.43295031008X^1 + 29.3064047061
Total error: 16335.123165

Newton's Method:
Fitting line: 4.43295031008X^1 + 29.3064047061
Total error: 16335.123165
Case 2: n = 3, =0

LSE:
Fitting line: 3.02385339349X^2 + 4.90619026386X^1
-0.231401756088
Total error: 26.5599594993

Newton's Method:
Fitting line: 3.02385339349X^2 + 4.90619026386X^1
-0.231401756088
Total error: 26.5599594993
Case 3: n = 3, = 10000

LSE:
Fitting line: 0.8345332827X^2 + 0.0931481983192X^1
+ 0.0469506992735
Total error: 22649.738493

Newton's Method:
Fitting line: 3.02385339349X^2 + 4.90619026386X^1
-0.231401756088
Total error: 26.5599594993

You might also like