C++ ML Playground: Linear Regression & Neural Network

A full-stack application demonstrating Linear Regression and a basic Feedforward Neural Network trained using C++ and visualized with React, Node.js, and WebSockets.

Features

• Linear Regression:
  • Train a model using the analytical solution (normal equation).
  • Visualize data points and the resulting regression line.
  • Make predictions on new data points.
  • View model performance metrics (MSE, R² score).
• Neural Network:
  • Define custom network structures (e.g., "1-4-1").
  • Train a feedforward network using gradient descent.
  • Specify learning rate and number of epochs.
  • Real-time training loss visualization (MSE vs. Epoch) via WebSockets.
  • Visualize NN predictions (orange triangles) alongside original data points, connected by a smooth interpolated line.
  • View final MSE and training time.
• General:
  • Input data manually or generate random datasets with adjustable linearity.
  • Interactive charts powered by Chart.js.
  • Modern UI built with React and DaisyUI/Tailwind CSS.
  • Dark/light theme toggle.

Project Structure

• client/: React frontend using create-react-app. Handles UI, visualization, and WebSocket communication.
• server/: Node.js/Express backend API. Manages requests, invokes the C++ executable, and relays NN training progress via WebSockets.
• cpp/: C++ engine containing:
  • linear_regression.h/.cpp: Implementation of the Linear Regression model.
  • neural_network.h/.cpp: Implementation of the Feedforward Neural Network.
  • main_server.cpp: Main C++ application handling command-line arguments (lr_train, nn_train_predict) and interacting with the Node.js server via stdin/stdout.
  • Makefile: Used to build the C++ executable.

Installation

Prerequisites

• Node.js (v14 or higher recommended)
• npm (usually comes with Node.js)
• A C++ compiler supporting C++11 or later (e.g., g++, Clang, MSVC)
• make build tool (standard on Linux/macOS, can be installed on Windows e.g., via Chocolatey choco install make or MinGW/MSYS2)

Setup

1. Clone the repository:

   git clone <your-repository-url>
   cd <repository-directory>

2. Build the C++ executable: The server expects the executable linear_regression_app (or linear_regression_app.exe on Windows) in the cpp/ directory.

   cd cpp
   make # This should create the 'linear_regression_app' executable
   cd ..

   (Note: Verify the Makefile target name matches linear_regression_app)

3. Install Server Dependencies and Start:

   cd server
   npm install
   npm start

   (Keep this terminal running)

4. Install Client Dependencies and Start: Open a new terminal in the project root.

   cd client
   npm install
   npm start

The application should now open automatically in your browser, typically at http://localhost:3000. The server runs on http://localhost:3001.

Usage

1. Input Data:

   • Enter comma-separated X and Y values in the text areas.
   • Alternatively, configure the number of points and linearity factor, then click "Generate" to create random data.

2. Linear Regression:

   • Click "Train Linear Regression".
   • Results (Slope, Intercept, MSE, R², Time) will appear below the button.
   • The regression line will be drawn on the chart.
   • Enter an X value under "Predict Y for LR Model" and click "Predict" to see the predicted point (green cross) and value.

3. Neural Network:

   • Configure the "Layer Sizes" (e.g., 1-4-1), "Learning Rate", and "Epochs".
   • Click "Train NN & Predict".
   • Training progress (MSE vs. Epoch) will stream to the loss chart below the button.
   • Once complete, final results (Final MSE, Time) will appear.
   • NN predictions for the input X values will be plotted as orange triangles on the main chart.

4. Visualization:

   • The main chart displays data points, the LR line (if trained), LR predictions (if made), and NN predictions (triangles connected by an interpolated curve, if trained).
   • Hover over points/lines for details.
   • The chart automatically adjusts its axes to fit the data and predictions.

Technologies

• Frontend: React, Chart.js, DaisyUI, Tailwind CSS
• Backend: Node.js, Express, ws (for WebSockets)
• C++ Engine: Standard C++ (C++11) for Linear Regression (analytical solution) and Neural Network (gradient descent).
• Build Tools: Make, npm

License

MIT

Acknowledgments

• DaisyUI for the UI components
• Chart.js for data visualization