Autoencoder for Dimensionality Reduction on MNIST

This project demonstrates how to use a simple autoencoder to perform dimensionality reduction on the MNIST dataset. The autoencoder is built with TensorFlow/Keras and learns to compress the 784-dimensional images into a lower-dimensional representation.

Project Overview

Autoencoders are a type of neural network used for unsupervised learning. They consist of two main parts:

1. Encoder: Compresses the input data into a lower-dimensional latent space.
2. Decoder: Reconstructs the original data from the compressed representation.

By training the autoencoder to minimize the reconstruction error, the encoder learns a meaningful, compressed representation of the data. This makes autoencoders a powerful tool for dimensionality reduction, especially for complex, nonlinear data.

Features

• Modular Code: The script is organized into functions for data loading, model building, training, and visualization.
• Hyperparameter Tuning: Easily experiment with hyperparameters like encoding_dim, epochs, and batch_size using command-line arguments.
• Visualization: The script generates two plots:
  • A 2D scatter plot of the encoded data, which helps visualize the learned latent space.
  • A comparison of original and reconstructed images to evaluate the model's performance.
• Model Saving: The trained autoencoder and encoder models are saved to the models/ directory for future use.

Getting Started

Prerequisites

• Python 3.x
• TensorFlow
• NumPy
• Matplotlib

Installation

1. Clone the repository:

   git clone https://github.com/your-username/autoencoder-mnist.git
   cd autoencoder-mnist

2. Create a virtual environment (optional but recommended):

   python -m venv venv
   source venv/bin/activate  # On Windows, use `venv\Scripts\activate`

3. Install the dependencies:

   pip install -r requirements.txt

Usage

You can run the script with default parameters:

python autoencoder_mnist.py

This will train the autoencoder for 20 epochs with an encoding dimension of 32. The script will then display the visualization plots and save the trained models to the models/ directory.

To experiment with different hyperparameters, you can use the following command-line arguments:

python autoencoder_mnist.py --encoding_dim 16 --epochs 30 --batch_size 128

• --encoding_dim: The dimension of the encoded representation (default: 32).
• --epochs: The number of epochs to train for (default: 20).
• --batch_size: The batch size for training (default: 256).

Output

The script will generate the following files:

• models/autoencoder.h5: The trained autoencoder model.
• models/encoder.h5: The trained encoder model.
• encoded_data_projection.png: A 2D scatter plot of the encoded test data.
• reconstructed_images.png: A comparison of original and reconstructed images.

License

This project is licensed under the MIT License. See the LICENSE file for details.

Autoencoder-for-Dimensionality-Reduction-on-MNIST