lineur - linear neuron (C++)

A lightweight C++ project implementing standalone single-layer neural models for:

• Binary classification
• Nonlinear regression

The project includes dataset generation, training visualization, and an interactive UI built with ImGui.

---

Overview

This project focuses on implementing neural models from scratch without external ML frameworks.
It is designed to clearly demonstrate the behavioral differences between classification and regression while providing real-time visualization of training dynamics.

Two independent neuron implementations are provided:

• ClNeuron – binary classification
• RegNeuron – regression

Each model has its own data structures and training logic.

---

Features

Classification

• Linear binary classifier
• Two bias strategies:
  • Explicit bias term (w0)
  • Threshold-based decision
• Custom dataset loading
• Interactive visualization of decision boundary

Regression

• Single-layer regression neuron
• Nonlinear target functions (e.g., parabola-based datasets)
• Optional noise injection in dataset generation
• Real-time prediction curve rendering

UI

• Built with ImGui
• Toolbar with:
  • Model selector (combo box)
  • Training controls
  • Dataset loading
• Interactive canvas:
  • Zoom
  • Mouse panning
  • Data point rendering
  • Model output rendering

---

Build

Requirements

• C++20 or newer
• CMake
• ImGui
• X11 (Linux)

Build Steps

mkdir build
cmake -S . -B build
cmake --build build

---

Dataset Format

Classification

x y label

Example:

 0.5 1.2 1
-0.3 0.8 0

Regression

x y

Example:

-2.0 4.1
 0.5 0.3