Vicentin

A comprehensive Python library for mathematical optimization, deep learning, computer vision, and classic algorithms. This library is designed with a dual-backend architecture, offering seamless switching between NumPy for transparency and PyTorch for hardware acceleration and automatic differentiation.

Table of Contents

• Introduction
• Features
• Installation
• Usage
• Pre-commit Setup
• License

Introduction

vicentin is a Python package that contains my personal implementations of a variety of algorithms, data structures, and optimization techniques. It serves as a collection of theoretical and practical programming concepts.

Features

• Data Structures: Queue, Stack, Tree, Graph, Heap, Priority Queue, Trie, Union Find
• Dynamic Programming: Knapsack, Matrix Multiplication, Rod Cutting, Edit Distance
• Graph Algorithms: Minimum Spanning Tree (MST), Shortest Path, Negative Cycle Detection
• Image & Video Processing: Optical Flow (Horn-Schunck), Differentiation, Regularization, Image-to-Graph
• Optimization: Gradient Descent, Newton's Method, Barrier Method, Proximal Gradient Descent, ISTA, Projected Gradient Descent, Newton-Raphson (Root Finding)
• Machine Learning:
  • Kernels
  • Unsupervised Learning: PCA
  • Classification: Logistic Regression, SVM
  • Regression: Ridge Regression
• Deep Learning:
  • Models: Autoencoders (AE), Variational Autoencoders (VAE)
  • Trainers: Standard, Supervised, GAN, Distillation
  • Losses: Beta-VAE, Wasserstein GAN
• Sorting: Heap Sort
• Mathematical Tools: Polynomial Operations
• Papers: Grad-CAM, SimCLR

Installation

pip install vicentin             # core: NumPy backend (numpy, scipy)
pip install "vicentin[torch]"    # + Torch backend for the dual-backend modules

Optional extras, combine as needed (e.g. "vicentin[torch,plot]"):

For development:

git clone https://codeberg.org/fsvicentin/vicentin.git
cd vicentin
python -m venv venv && source venv/bin/activate
pip install -e ".[test]"
pytest

Usage

Heap data structure

# Example: Using the heap data structure
from vicentin.data_structures.heap import Heap

heap = Heap()
heap.insert(5)
heap.insert(2)
heap.insert(8)

print(heap.extract_min())  # Output: 2

Newton's Method

import torch
from vicentin.optimization.minimization import newton_method

def objective(x):
    return torch.sum(x**2)

x0 = torch.tensor([10.0, 10.0])
A = torch.tensor([[1.0, 1.0]])
b = torch.tensor([1.0])

# Backend (Torch) is automatically detected from x0
x_opt = newton_method(objective, x0, equality=(A, b))
print(f"Optimal solution: {x_opt}")

Pre-commit Setup

This repository uses pre-commit to enforce coding standards, automatic formatting and automatic version bumping before commits.

1️⃣ Install pre-commit

pip install pre-commit

2️⃣ Install Hooks

pre-commit install

3️⃣ Use commitizen to commit

cz commit

License

This project is licensed under the MIT License.