Neural Network Project

A simple neural network project using Node.js.
This project allows you to train a small neural network and run it interactively via the console.

Features

• Load training data from trainingData.json
• Create or load a neural network (net.json)
• Save network memory (memory.json)
• Configurable network architecture and training parameters via config.json
• Interactive mode to test network predictions

Files

| File | Purpose | |------|---------| | index.js | Main script: trains network (if needed) and starts interactive mode | | nn.js | Neural network implementation | | config.json | Configuration file for network and training parameters | | .gitignore | Ignore local data and node modules | | package.json | Project metadata and scripts |

Intallation

npm install smallnn

Installation with Git

1. Clone the repository:

git clone https://github.com/LeeJiwooKor/smallnn
cd smallnn

2. Install Node.js dependencies (if any; currently none required):

npm install

Configuration

config.json controls network parameters and file paths. Example:

{
  "trainingDataFile": "trainingData.json",
  "netFile": "net.json",
  "memoryFile": "memory.json",
  "layers": [
    { "nodes": 2 },
    { "nodes": 8, "activation": "relu" },
    { "nodes": 1, "activation": "sigmoid" }
  ],
  "learningRate": 0.05,
  "iterations": 5000
}

You can modify:

• layers: number of layers, nodes per layer, activation functions
• learningRate: speed of learning
• iterations: number of training cycles
• file paths: names of training data, network, and memory files

Usage

1. Make sure trainingData.json exists in your project folder. Format:

[
  { "input": [0, 0], "output": [0] },
  { "input": [0, 1], "output": [1] },
  { "input": [1, 0], "output": [1] },
  { "input": [1, 1], "output": [0] }
]

2. Run the script:

node index.js

3. The script will:

• Load config.json and training data
• Load existing network (net.json) or create a new one
• Load memory (memory.json) if it exists
• Train the network if needed
• Start interactive mode to input values and see output

4. Example input in interactive mode:

Enter input (comma separated, e.g. 1,0): 0,1
Output: [0.987654]

Notes

• Do not commit large files like trainingData.json, net.json, or memory.json—keep them local.
• You can reset the network by deleting net.json and memory.json.
• The neural network currently supports a small feedforward setup; modify nn.js for more advanced features.

Optional: npm scripts

"scripts": {
  "start": "node index.js"
}

Run:

npm start

License

MIT License