Scribd
Upload
76 views3 pages

Question 2

This document contains the results of various experiments with K-means clustering and support vector machines. For K-means, it reports the sum of squares and cluster assignments for different values of K. For SVMs, it shows that grid search tuning of the C and gamma hyperparameters significantly improved cross validation accuracy from 15.6% to over 90%. It also reports the best hyperparameters found and test set accuracy of 76.8%.

Uploaded by

JaspreetSingh
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
76 views3 pages

Question 2

This document contains the results of various experiments with K-means clustering and support vector machines. For K-means, it reports the sum of squares and cluster assignments for different values of K. For SVMs, it shows that grid search tuning of the C and gamma hyperparameters significantly improved cross validation accuracy from 15.6% to over 90%. It also reports the best hyperparameters found and test set accuracy of 76.8%.

Uploaded by

JaspreetSingh
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/ 3

Question 2

Question 2.5.1
K: 2
Sum of squares: 8.430164e+05 p1: 77 p2: 46 p3: 6.150000e+01

K: 4
Sum of squares: 7.663217e+05 p1: 54 p2: 84 p3: 69

K: 6
Sum of squares: 7.399330e+05 p1: 45 p2: 92 p3: 6.850000e+01

Question 2.5.2
K: 6
Iterations: 4

Question 2.5.3

Plots made on average of 25 runs.


Question 2.5.4

Question 3

Question 3.4.2
Default LibSVM parameters
Cross Validation Accuracy = 15.6443%

Question 3.4.3
Grid search was performed to tune C and gamma (RBF kernel),
Below is the combination of parameters showing highest cross validation accuracy:
C: 100 G: 1 Cross Validation Accuracy = 92.009%
C: 10000 G: 1.0e-03 Cross Validation Accuracy = 90.9398%

For full list see matlab script output


For G=1 and larger C values Cross Validation Accuracy remains high and same.
Also for larger C values even small gamma produced good Cross Validation
Accuracy.
Question 3.4.5
The estimated value of gamma for kernel was: 20.63794
Hence search was done in order of values of tens

Again grid search was used for kernel parameter gamma and SVM
parameter C for tuning
Below is the one of combination with best Cross Validation Accuracy.
C: 22 G: 50 Cross Validation Accuracy = 93.4159%

For G in range 25 to 75 gives high Cross Validation Accuracy


with large enough C value, like greater than 20

Question 3.4.6
Accuracy on test set: 76.8%

You might also like