MNIST Neural Network Explorer

An interactive web application for exploring and understanding neural networks through the MNIST dataset. Train a neural network in your browser and watch it learn to recognize handwritten digits in real-time.

Features

• Interactive Dataset Exploration

  • Browse through the MNIST training dataset in single or grid view
  • Examine individual digits with adjustable size
  • View 100 digits at once in a 10x10 grid layout

• Neural Network Training

  • Train a 3-layer neural network directly in your browser
  • Customize training parameters (epochs, learning rate, batch size, hidden nodes)
  • Real-time training progress and accuracy visualization
  • Network architecture visualization showing weights and connections

• Model Testing

  • Draw digits using mouse or touch input
  • Test against real MNIST test images
  • View prediction confidence for each digit
  • Analyze model performance through confusion matrix

Installation and Setup

1. Clone the repository:

   $ git clone https://github.com/malminhas/friendly-digits-explorer.git
   $ cd friendly-digits-explorer

2. Install dependencies:

   $ npm install

3. Run the development server:

   $ npm run dev

4. Open http://localhost:8081 in your browser

Project Structure

src/
├── components/           # React components
│   ├── explore/         # Dataset exploration components
│   ├── ui/             # Reusable UI components
│   └── ...             # Main feature components
├── context/            # React context providers
├── hooks/             # Custom React hooks
├── lib/               # Core functionality
│   ├── mnist.ts       # MNIST data loading and processing
│   └── neural-network.ts # Neural network implementation
└── pages/             # Next.js pages

Architecture

The application follows a modular architecture with clear separation of concerns:

---
title: MNIST Neural Network Explorer Architecture
---
graph TB
    subgraph Browser
        UI[User Interface]
        NNC[Neural Network Context]
        NN[Neural Network Engine]
        Canvas[Canvas Drawing]
        
        subgraph Data Processing
            MNIST[MNIST Data Loader]
            Preprocess[Image Preprocessor]
        end
        
        subgraph Storage
            LocalStore[Local Storage]
            Cache[Browser Cache]
        end
    end
    
    subgraph External
        CDN[MNIST Dataset CDN]
    end
    
    %% Data flow
    CDN -->|Load Dataset| MNIST
    MNIST -->|Process| NNC
    UI -->|User Input| Canvas
    Canvas -->|Process Image| Preprocess
    Preprocess -->|Forward Pass| NN
    NN -->|Update Weights| NNC
    NNC -->|Store Model| LocalStore
    NNC -->|Update UI| UI
    
    %% Styling
    classDef primary fill:#4338ca,stroke:#4338ca,color:#fff
    classDef secondary fill:#6366f1,stroke:#6366f1,color:#fff
    classDef storage fill:#059669,stroke:#059669,color:#fff
    classDef external fill:#dc2626,stroke:#dc2626,color:#fff
    
    class UI,Canvas primary
    class NNC,NN,MNIST,Preprocess secondary
    class LocalStore,Cache storage
    class CDN external

Loading

Component Interactions

1. Data Flow

   • MNIST dataset is loaded from CDN and processed by the Data Loader
   • Neural Network Context manages state and model parameters
   • User interactions trigger canvas updates and predictions
   • Model state is persisted in Local Storage

2. Key Components

   • UI Layer: React components for user interaction
   • Neural Network Context: Central state management
   • Processing Layer: Image preprocessing and neural network computations
   • Storage Layer: Model persistence and data caching

3. Performance Optimizations

   • Browser caching for MNIST dataset
   • Local storage for trained model persistence
   • Efficient canvas operations for drawing
   • Optimized matrix operations for network computations

Neural Network Implementation

The application implements a 3-layer feedforward neural network:

• Input Layer: 784 nodes (28x28 pixels)
• Hidden Layer: Configurable (default 128 nodes)
• Output Layer: 10 nodes (digits 0-9)

Key Features:

• Forward Propagation: Uses ReLU activation for hidden layer and softmax for output
• Backward Propagation: Implements stochastic gradient descent
• Training Process:
  • Runs entirely in JavaScript on the client-side CPU
  • Uses mini-batch processing for efficiency
  • Implements cross-entropy loss function
  • Supports customizable learning rates and batch sizes

Data Processing:

• Normalizes pixel values to [0,1] range
• Implements proper centering and scaling of drawn digits
• Uses anti-aliasing for better quality digit recognition

Performance

• Training Data: Uses a subset of 3,000 training images and 1,000 test images for faster loading
• Training Time: ~2-3 minutes for 30 epochs on modern browsers
• Model Size: ~400KB when exported (weights + biases)
• Recognition Accuracy:
  • ~92-95% on MNIST test set
  • ~85-90% on hand-drawn digits

Optimization Techniques:

• Progressive data loading
• Efficient matrix operations
• Canvas optimization for drawing
• Web Workers for background processing (coming soon)

Deployment

The application can be deployed either locally on your machine or remotely on a DigitalOcean droplet using Terraform.

Local Deployment (Serve at /)g

• Docker installed on your machine
• Terraform installed (for deployment)
• SSH key pair (for remote deployment)
• DigitalOcean droplet (for remote deployment)

Local Deployment (Serve at /)

You can deploy locally using either the default terraform.tfvars or by passing variables on the command line.

Option 1: Using terraform.tfvars

1. Ensure your terraform/terraform.tfvars file contains:

   droplet_ip = "your-droplet-ip"
   private_key_path = "path/to/your/private/key"
   container_name   = "friendly-digits-explorer"
   container_port   = 8081
   host_port        = 8081
   vite_base = "/"
   vite_basename = "/"
   environment = "local"
   # Optionally, for Apple Silicon:
   # build_platform = "linux/arm64"

2. From the terraform directory, run:

   terraform apply -auto-approve

Option 2: Passing variables on the command line

cd terraform
terraform apply -auto-approve \
  -var="vite_base=/" \
  -var="vite_basename=/" \
  -var="environment=local"

• Access the app at: http://localhost:8081

---

Remote Deployment (Serve at /friendly-digits-explorer)

You can deploy remotely using either the default terraform.tfvars or by passing variables on the command line.

Option 1: Using terraform.tfvars

1. Ensure your terraform/terraform.tfvars file contains:

   droplet_ip = "your-droplet-ip"
   private_key_path = "path/to/your/private/key"
   container_name   = "friendly-digits-explorer"
   container_port   = 8081
   host_port        = 8081
   vite_base = "/friendly-digits-explorer/"
   vite_basename = "/friendly-digits-explorer"
   environment = "remote"
   # Optionally, for Apple Silicon:
   # build_platform = "linux/arm64"

2. From the terraform directory, run:

   terraform apply -auto-approve

Option 2: Passing variables on the command line

cd terraform
terraform apply -auto-approve \
  -var="vite_base=/friendly-digits-explorer/" \
  -var="vite_basename=/friendly-digits-explorer" \
  -var="environment=remote"

• Access the app at: https://yourdomain.com/friendly-digits-explorer

Caddyfile for subpath proxying:

foobar.co.uk {
    handle_path /friendly-digits-explorer/* {
        reverse_proxy localhost:8081
    }
}

---

Terraform Commands Reference

• terraform init: Initialize Terraform and download required providers
• terraform plan: Show the execution plan without making changes
• terraform apply: Apply the changes to create/update resources
• terraform destroy: Remove all resources created by Terraform
• terraform state list: List all resources in the current state
• terraform state show <resource>: Show details of a specific resource

Deployment Workflow

The Terraform configuration:

1. Builds the Docker image locally using buildx
2. Saves the image to a tar file (for remote)
3. Copies the tar file to the remote server (for remote)
4. Loads the image on the remote server (for remote)
5. Starts the container with the specified configuration

Troubleshooting

1. If the container fails to start, check the logs:

   $ docker logs friendly-digits-explorer

2. If the image fails to load on the remote server:

   $ ssh -i your-key root@your-droplet-ip "docker images"

3. To clean up and start fresh:

   $ terraform destroy -auto-approve

Customizing the Build Platform

You can control which architecture the Docker image is built for using the build_platform variable in terraform.tfvars or via the command line. This is useful if you want to run the container natively on your Mac (Apple Silicon) or for remote x86 servers.

• Apple Silicon Mac (M1/M2/M3):

  $ terraform apply -var="environment=local" -var="build_platform=linux/arm64" -auto-approve

• Intel Mac or remote/x86:

  $ terraform apply -var="environment=local" -var="build_platform=linux/amd64" -auto-approve

• Default: If you do not specify, it will use linux/amd64.

You can also set this in your terraform/terraform.tfvars file:

build_platform = "linux/arm64" # or "linux/amd64"

---

Notes

• You can override any variable in terraform.tfvars via the -var command line flag.
• The vite_base and vite_basename variables control the asset and router base paths for your build.
• For local, always use /. For remote subpath, use /friendly-digits-explorer/ and /friendly-digits-explorer.

Contributing

Contributions are welcome! Here's how you can help:

1. Fork the repository
2. Create a feature branch: git checkout -b feature/amazing-feature
3. Commit your changes: git commit -m 'Add amazing feature'
4. Push to the branch: git push origin feature/amazing-feature
5. Open a Pull Request

Areas for Improvement:

• GPU acceleration support
• Additional network architectures
• Extended dataset support
• Performance optimizations
• Mobile responsiveness
• Accessibility improvements

License

This project is licensed under the MIT License - see below for details:

MIT License

Copyright (c) 2024 [Your Name]

Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all
copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
SOFTWARE.