Machine Learning Assignments

Assignments for the Introduction to Machine Learning course (26VT-2DV516) at Linnaeus University (LNU).

Learning	Assignment 1	Assignment 2	Assignment 3	Exam Drills
just for fun idk	Grade: A	Grade: A	Grade: B	/web-exam

Project Structure

.
├── learning.ipynb / learning.py   # Introductory Python, NumPy, Matplotlib, and scikit-learn notebook
├── __marimo__/                    # Marimo session state
├── pyproject.toml                 # Project metadata, dependencies, and Poe tasks
├── uv.lock                        # Locked dependency versions
├── .python-version                # Python 3.14
│
├── assignments/
│   ├── assignment-1/
│   │   ├── A1.ipynb / A1.py       # k-Nearest Neighbors
│   │   ├── __marimo__/            # Marimo session state
│   │   ├── MachineLearningModel.py
│   │   ├── IrisDataset.csv
│   │   ├── Polynomial200.csv
│   │   └── mnist/                 # MNIST handwritten digits dataset
│   │
│   ├── assignment-2/
│   │   ├── pp222rp_A2.ipynb / pp222rp_A2.py
│   │   ├── __marimo__/            # Marimo session state
│   │   ├── MachineLearningModel.py
│   │   ├── DecisionBoundary.py
│   │   ├── ForwardSelection.py
│   │   ├── ROCAnalysis.py
│   │   └── datasets/              # Banknote, heart disease, Boston housing, secret polynomial
│   │
│   └── assignment-3/
│       ├── pp222rp_A3.ipynb / pp222rp_A3.py
│       ├── __marimo__/            # Marimo session state
│       ├── bkmeans.py             # Bisecting k-Means implementation
│       ├── datasets/
│       │   ├── BankMarketing/     # Bank Marketing dataset
│       │   └── your_datasets/     # Adult, default, and dropout datasets
│   └── *.png                  # Generated plots used by the assignment
│
├── web-exam/                      # React SPA — exam drill questions with simulations & self-grading
│
├── datasets/                      # Shared datasets (admission, Boston housing, iris, microchips, secret polynomial)
└── docs/
    ├── __marimo__/                # Marimo session state
    └── pyplot.ipynb / pyplot.py   # Matplotlib notes/tutorial

Assignments

Assignment 1: k-Nearest Neighbors

Implementation of k-NN regression and classification from scratch with NumPy, Matplotlib, and SciPy.

• KNN regression and classification: custom models using Euclidean distance, tested on Polynomial200 and Iris with decision-boundary contour plots across varying k.
• Regression validation: repeated random-split experiments with MSE bar charts across multiple k values.
• Feature combination search: exhaustive search over 2-feature combinations on Iris to find the best classification performance.
• Fast KNN with KDTree: SciPy KDTree-based model benchmarked for performance against the naive implementation.
• MNIST classification: handwritten-digit classification on MNIST using the fast KNN model.

Assignment 2: Regression, Classification, and Neural Networks

Implementation of regression and classification models from scratch, followed by scikit-learn experiments.

• Normal equation and gradient descent regression: closed-form and iterative polynomial regression on Boston Housing, with normalization, cost-evolution plots, and hyperparameter tuning.
• Polynomial degree selection: model-fit comparison across polynomial degrees with shuffled-run validation.
• Logistic and non-linear logistic regression: binary classification on banknote authentication, hyperparameter analysis, and decision boundary visualizations.
• ROC analysis and forward selection: custom ROC metrics and feature selection on the heart disease Cleveland dataset.
• Neural networks: MLPClassifier with GridSearchCV on the digits dataset, including confusion matrix and loss analysis.

Assignment 3: Trees, Ensembles, SVM, PCA, and Clustering

Use of scikit-learn and custom algorithms for advanced machine-learning tasks.

• Decision trees and ensembles: Bank Marketing dataset preprocessing, tree visualization, depth/overfitting analysis, and comparison of Random Forest and Gradient Boosting with ROC/AUC and feature importance plots.
• Robustness to noise: synthetic noise injected into the dataset to test ensemble model resilience.
• SVM: Breast Cancer dataset with PCA, linear vs RBF kernel comparison, grid search over C and gamma, and logistic regression benchmark.
• Bisecting k-Means: custom implementation from scratch, tested on synthetic blobs and 3 real-world datasets with PCA projections.
• Dimensionality reduction and clustering comparison: PCA, MDS, and t-SNE side by side, plus comparison of bisecting k-Means, classic k-Means, and hierarchical clustering on t-SNE projections.

Installation

This project requires Python 3.14 and uv.

git clone https://github.com/TeenBiscuits/Practicas-ML.git
cd Practicas-ML
uv sync

Running Notebooks

Run commands from the repository root so the relative dataset paths used by the notebooks resolve correctly.

# Open the Marimo editor
uv run poe marimo

# Open Jupyter Lab
uv run poe lab

You can also open the notebooks directly from GitHub or Molab using the badges above.

Technologies

• Python 3.14
• uv for dependency management
• Marimo and Jupyter Lab for notebooks
• NumPy for array operations and linear algebra
• Matplotlib for visualizations
• scikit-learn for models, metrics, preprocessing, dimensionality reduction, and clustering
• SciPy for KDTree-based nearest-neighbor search
• React + TypeScript + Tailwind CSS + Vite for the exam practice web app