Reference no: EM133032239

Assignment: Support Vector Machine Regressor

This assignment mixes theory and application, in the form of two problems. Perform the tasks described in each. Note that this (and other) assignments includes a few challenging research-related tasks. They are aimed at gradually building your capacity to tackle complex topics, familiarize yourself with academic discourse, and provide context and practice for the skills you will eventually need when working on your capstone thesis or project.

Part 1 - Theory
You have a choice of handling a binary classification task using number of misclassifications as the performance measure and maximizing the margin between the two classes as the performance measure. On what factors does your decision depend? Provide a formal explanation, supported by theorems and ideas presented in the readings associated with this topic.

You have a choice of handling a binary classification task using (i) linear SVM, and (ii) perceptronalgorithm. On what factors does your decision depend? Provide a formal explanation, supported by theorems and ideas presented in the readings associated with this topic.

Part 2 - Application
The sinc function is one of the commonly used datasets for testing nonlinear regression algorithms. This function is given by the following equation:

sinc(x)=sinπx/πx

Familiarize yourself with the SVM tools in Python, which can be found within your topic materials.

Create a Jupyter notebook and implement the following (in Python):

Generate 50 data points from this function in the range [- 3, 3].
Add Gaussian noise to the data.
Train an SVM regressor with the data generated in (a). Define (and explain) suitable parameters required for training the regressor.
Describe the functionality of the regressor.
Discuss the potential use of the regressor and quantify its accuracy.

After you assess the importance and approach to using the sinc function in conjunction with SVM, refer to "A Signal Theory Approach to Support Vector Classification: The Sinc Kernel" within your topic materials. You are not expected to grasp all the concepts and theorems described in the article, but skim through it on a high level, to get some insight into the work it describes. Upon skimming through the article, expand your discussion in (e) above, to include some of the relevant points.

Article - A signal theory approach to support vector classification: The sinc kernelI James D.B. Nelson, Robert I. Damper∗, Steve R. Gunn, Baofeng Guo