COMP527: Data mining and visualisation - CA Assignment 1: Data Classification - Implementing Perceptron algorithm
CA Assignment 1 Data Classification
Implementing Perceptron algorithm
Assessment Information
| Assignment Number | 1 (of 2) |
|---|---|
| Weighting | 15% |
| Assignment Circulated | 10 Feb 2023 |
| Submission Mode | Electronic via Canvas |
| Purpose of assessment | The purpose of this assignment is to demonstrate: (1) the understanding of the Perceptron agorithm; (2) the ability to implement the Perceptron algorithm for binary classification; (3) the ability to evaluate a classification algorithm; (4) the ability to turn a binary classification algorithm to a multi-class classification algorithm using the 1-vs-rest approach; (4) the ability to incorporate regularisation into a classification algorithm. |
| Learning outcome assessed | (1) A critical awareness of current problems and research issues in data mining. (3) The ability to consistently apply knowledge concerning current data mining research issues in an original manner and produce work which is at the forefront of current developments in the sub-discipline of data mining. |
Objectives
This assignment requires you to implement the Perceptron algorithm using the Python programming language.
Assignment description
Download the CA1data.zip file. Inside, you will find two files: train.data and test.data, corresponding respectively to the train and test data to be used in this assignment. Each line in the file represents a different train/test instance. The first four values (separated by commas) are feature values for four features. The last element is the class label (class-1, class-2 or class-3).
Questions/Tasks
- (15 marks) Explain the Perceptron algorithm (both the training and the test procedures) for the binary classification case. Provide the pseudo code of the algorithm. It should be the most basic version of the Perceptron algorithm, i.e. the one that was discussed in the lectures.
- (30 marks) Implement a binary perceptron. The implementation should be consistent with the pseudo code in the answer to Question 1.
- (15 marks) Use the binary perceptron to train classifiers to discriminate between • class 1 and class 2, • class 2 and class 3, and • class 1 and class 3. Report the train and test classification accuracies for each of the three classifiers after training for 20 iterations. Which pair of classes is most difficult to separate?
- (30 marks) Explain in your own words what the 1-vs-rest approach consist of. Extend the binary perceptron that you implemented in part 3 above to perform multi-class classification using the 1-vs-rest approach.
- (10 marks) Add an `2 regularisation term to your multi-class classifier implemented in part 4. Set the regularisation coefficient to 0.01, 0.1, 1.0, 10.0, 100.0 and compare the train and test classification accuracies. What can you conclude from the results?
Submission Instructions
Submit via Canvas the following two files (please do NOT zip files into an archive)
- the source code for all your programs (do not provide ipython/jupyter/colab notebooks, instead submit standalone code in a single .py file), and
- a PDF file (report) of no more than 3 pages providing the answers to the questions. It is extremely important that you provide the two files described above and not just the source code!
Important notes
(read carefully and double check compliance before submission)
- No credit will be given for implementing any other type of classification algorithm or using an existing library for classification instead of implementing it by yourself. However, you are allowed to use • numpy library for accessing data structures such as numpy.array; • random module; and • pandas.read_csv, csv.reader, or similar modules only for reading data from the files. However, it is not a requirement of the assignment to use any of those modules.
- Your program • should run and produce all results for Questions 3, 4, and 5 in one click without requiring any changes to the code;
- Programs that do not run will result in a mark of zero!
- Your code should be as clear as possible and should contain only the functionality needed to answer the questions. Provide as much comments as needed to make sure that the logic of the code is clear enough to a marker. Marks may be deducted if the code is obscure, implements unnecessary functionality, or is overly complicated.
- You are allowed to shuffle the data. If you use module random to shuffle the data, use a fixed seed value so that your program always produces the same output. This output should be exactly the one that you provide in the PDF report.
- Your answers in the PDF report should be succinct, but complete and clear. The clarity and presentation of the report will be assessed.
