Homework answers / question archive / CSE 6363: Machine Learning University of Texas at Arlington Summer 2022 Assignment 3 This assignment uses Support Vector Machines

CSE 6363: Machine Learning University of Texas at Arlington Summer 2022 Assignment 3 This assignment uses Support Vector Machines

Statistics

Share With

CSE 6363: Machine Learning

University of Texas at Arlington

Summer 2022

Assignment 3

This assignment uses Support Vector Machines.

(20 points) The Negative Case
Given the code on https://github.com/ajdillhoff/CSE6363/blob/main/svm/smo.ipynb and
the pseudocode given in the original Platt paper, implement the case when is negative.
(40 points) Non-linear SVM
Adapt the given code to support non-linear SVMs. Modify the SVM class given so that
if the input for kernel is poly, the SVM will use a polynomial kernel. To test this, use
sklearn.datasets.make_circles to generate a non-linear dataset and fit your SVM to
it. It should be able to correctly separate the two classes.
An example of how to generate the dataset is given here. Make sure to set aside 10% of
the samples as a testing set.
Compare your implementation with sklearn.svm.SVC using both a linear kernel and
polynomial kernel. Your implementation is not expected to perform better, but
should behave similarly.
In your code, print the accuracy result of the test set and visualize the output using
plot_decision_regions.
(40 points) Multi-class SVM
Add multi-class support to your implementation following a One-versus-All approach.
Given K classes, you will need to train K SVMs to classify one class versus all others
combined.
When implementing this, create another python class named MultiSVM which internally
represents the individual binary SVMs. The prediction function should classify the sample
according to which binary classifier gives the largest score.
Compare your implementation with sklearn.svm.SVC using both a linear kernel and polynomial
kernel. Use the Iris dataset provided by sklearn.datasets.load_iris. Your
implementation is not expected to perform better, but should behave similarly.

Submission
Create a zip file that includes all of your code as well as your report. The TA should
be able to easily run the code to reproduce all plots and results. Include any additional
instructions, if necessary.