UNIGE-Machine-Learning

[Master course] Lab assignments developed during the Machine Learning course at UNIGE

Requirements

• MATLAB
• LaTeX

Assignments summary

1. Linear classifier
   • No algorithms developed, instead a manual-visual approach was used to select the hyperplane dividing 2 linearly-separable classes.
2. Naive Bayes classifier
   • Developed naive bayes classifier on categorical data (multivariate multinomial distribution).
   • Dataset consisting of 14 observations with 4 features each.
   • Binary classification (yes, no)
3. k-Nearest Neighbors classifier
   • lazy-learner algorithm developed and tested on semeion-digits dataset (handwritten scanned digits).
   • Introduction to the confusion matrix and its terminology (accuracy, recall, precision, F1 score)
   • t-SNE embedding (optional visualization usng built-in MATLAB function)
   • Multiclass classification (0, 1, 2, ..., 9)
4. Percetron and cross-validation
   • Developed Rosenblatt's perceptron using sign as activation function.
   • Tested with semeion-digits dataset (handwritten digits)
   • Cross validation and hyper-parameter tuning.
   • Binary (one-vs-all) and multiclass classification.
   • Comparison between performance of perceptron and k-NN.
5. Autoencoder using NNs
   • Collaborative lab with Alexandre Sarazin
   • Use of MATLAB's nntool (neural networks toolbox)
   • Autoencoder using a simple Neural Netwrok with 1 hidden layer.

Klassifiers

(cleverly intended misspelling)

This folder contains the implementation of

• kNN classification algorithm
• Rosenblatt's perceptron

common-functions

Useful functions used throghout the reports

• Confusion matrix related
  • plot_confMat.m
  • accuracy.m
  • precision.m
  • recall.m
  • F1Score.m
  • specificity.m
• Cross-validation
  • cross_val_score.m
  • get_scores.m
  • merge_scores.m
• Train-test-split
  • train_test_split.m
  • stratified_split.m