/Digit-Recognition-MDX-Coursework

Primary LanguageRich Text Format

Digits Recognition Coursework

Coursework Overview

  • Subject: Machine Learning
  • Due Date: Week 19 (February 22nd, 2024)
  • Contribution to Overall Course Mark: 40%

Task Description

  • Develop a machine learning system from scratch.
  • Categorize one of the UCI digit tasks using the Optical Recognition of Handwritten Digits Data Set.
  • Use converted data sets: cw2 data set 1 and cw2 data set 2.
  • Perform a two-fold test and report the results.

Development Requirements

  • Code should be written in Java.
  • Must run in Eclipse.
  • If using existing algorithms, reference them in both the code and the report.

Marking Scheme (Total: 100 Points)

  1. Report (20 Points)
  2. Running Code (20 Points)
  3. Quality of Code (20 Points)
  4. Quality of Algorithm (20 Points)
  5. Quality of Results (20 Points)

Report Guidelines

  • Length: 1-2 pages.
  • Include a description of the chosen algorithm and the rationale behind this choice.
  • Show results of the two-fold test.
  • Discuss data usage briefly.

Coding and Algorithm Tips

  • Ensure the code is well-commented and structured.
  • A balance between simple and complex algorithms is crucial.
  • Aim for surpassing the baseline performance reported on the UCI website.
  • Starting with a nearest neighbour algorithm using Euclidean distance is suggested.

Submission Requirements

  • Submit code, a mark sheet, and analysis to the coursework folder of CST 3170 on myunihub.
  • Optionally, email a copy to the tutor.

Additional Notes

  • This is not a competition; discussing with colleagues can be beneficial.
  • The baseline for passing includes implementing a nearest neighbour algorithm with Euclidean distance.

Resources:

SVM Notes

  • The primary objective of SVM is to find a hyperplane that maximally separates the different classes in the dataset.
  • The "weight(w)" vector determines the orientation of the hyperplane, while the "bias(b)" term determines the offset of the hyperplane from the origin.

RBF Notes

  • K(x,x') = exp(-gamma||x-x'||^2)
  • Gamma is a hyperparameter that determines the spread of the Gaussian around the support vectors.
    • A small gamma value will result in a large variance and a small bias, while a large gamma value will result in a small variance and a large bias.
    • The gamma parameter can be thought of as the inverse of the radius of influence of samples selected by the model as support vectors.
    • Recommended: 0.5
  • C is the regularization parameter, which controls the trade-off between achieving a low training error and a low testing error that is the ability to generalize your classifier to unseen data.
    • A large C value will result in a small margin and a small number of support vectors, while a small C value will result in a large margin and a large number of support vectors.
    • Recommended: 0.01
  • Optimizer:

0.8424821002386634

  • Max_Itterations: 100000
  • Kernel: linear
  • C: 0.5
  • epsilon: 0.0001
  • weightThreshold: 0.0001
  • bias: 0.0
  • alpha: 0.00001
  • gamma: 0.001

0.9594272076372315

  • Max_Itterations: 100000000
  • Kernel: linear
  • C: 0.85
  • epsilon: 0.00001
  • weightThreshold: 0.00001
  • bias: 0
  • alpha: 0.000001
  • gamma: 0.5

All code commented, All classes with docstrings, no 1 letter vars.

Function Length < 50 lines.

2Fold test, use dataset 1 for test and dataset 2 for train and vice versa.

    int numClasses = 10; // for digits 0-9
    int MAX_ITERATIONS = 100000000;
    String kernelType = "linear";
    double C = 0.85;
    double epsilon = 0.00001;
    double weightThreshold = 0.00001;
    double bias = 0;
    double alpha = 0.000001;
    double gamma = 0.5;

Accuracy: 95.62277580071175%