Neural Network in C++ from Scratch

This project implements a basic feedforward neural network in C++ using custom layers, neurons, and backpropagation. The network can be trained on a binary classification dataset.

The project is organized as follows:

neunet/ 
  ├── src/
    │ ├── network/
      │ │ ├── net.h
      │ │ ├── net.cpp
      │ │ ├── layer.h
      │ │ ├── layer.cpp
      │ │ ├── neuron.h
      │ │ ├── neuron.cpp
    │ ├── optimizer/
      │ │ ├── utils.h
      │ │ ├── utils.cpp
    │ └── main.cpp
    ├── data.txt
    └── Build.bat

Prerequisites

• C++ Compiler: You need a C++ compiler like g++ to compile the project. You can install it as part of the MinGW toolset, available on Windows.
• Command Prompt (cmd): The project is set up to run in a Windows environment using cmd.

Setup and Compilation

Step 1: Install MinGW (if not already installed)

If you do not have g++ installed on your system, follow these steps:

1. Download the MinGW Installer from here.
2. During installation, select the g++ compiler.
3. Add the bin directory of your MinGW installation to your system’s PATH environment variable.

Step 2: Prepare the Dataset

You will need a dataset file (data.txt) that contains rows of data, with the last value of each row being the label.

Place the data.txt file in the root of the project directory.

Step 3: Compile the Project

1. Open the Command Prompt.

2. Navigate to the src/ directory of your project:

   cd path\to\your\project\src```

3. Run the provided Build.bat script to compile the project.

If the build is successful, you will see a neural_net.exe executable generated in the src/ folder.

Running the Program

Once the project has been compiled, follow these steps to run the neural network:

Step 1: Run the Executable

In the Command Prompt, ensure you are in the src/ directory. Run the following command:

neural_net.exe

Step 2: Input the Neural Network Configuration

The program will prompt you to enter various parameters for configuring the neural network:

• Number of Layers: Enter the total number of layers (excluding the input layer).
• Neurons per Layer: For each layer, input the number of neurons.
• Learning Rate: Input the learning rate (e.g., 0.01).
• Training Epochs: Input the number of epochs to train the network.

Step 3: Provide a Test Data Point

After training, the program will prompt you to enter a test data point (space-separated values). The test input should match the dimensionality of the input layer.\

Enter a test data point [space-separated]: 1.5 2.3 4.0

Project Structure:

• main.cpp: The entry point of the program. It initializes the neural network, loads the dataset, and manages training and prediction.
• network/: Contains the core implementation of the neural network.
• net.cpp: Manages the entire network, layers, and training.
• layer.cpp: Defines the layer class, which contains neurons and performs forward and backward propagation.
• neuron.cpp: Defines the neuron class, which holds weights, biases, and activation functions.
• optimizer/utils.cpp: Provides utility functions for loading the dataset and performing weight updates.

Sample Output:

Enter the layers in the neural network: 3
Enter the number of neurons in layer 1: 4
Enter the number of neurons in layer 2: 5
Enter the number of neurons in layer 3: 1
Enter the learning rate: 0.01
Enter the number of training epochs: 1000
Training the network...
Enter a test data point [space-separated]: 2.0 1.5 3.0
Predicted output: 0.85g

License

This project is licensed under the MIT License. You are free to modify and distribute it as you wish.