t-SNE and Spectral Clustering Project

This project demonstrates the use of dimensionality reduction and clustering techniques on image datasets for unsupervised learning and visualization.

---

📋 Overview

Key Techniques:

• t-SNE: t-Distributed Stochastic Neighbor Embedding for dimensionality reduction.
• Spectral Clustering: Unsupervised learning to group data into clusters.
• Graph Embedding: Techniques like Laplacian Eigenmaps, Kernel PCA, Isomap, and LLE.

Visualizations:

• High-dimensional data visualized in 2D and 3D spaces.
• Clustering results and performance metrics.

---

📂 Datasets

Supported Datasets:

1. MNIST: Handwritten digits dataset.
2. CIFAR-10: Color images across 10 categories.

---

✨ Features

• Data Preprocessing: Normalization and preparation for analysis.
• Dimensionality Reduction: Using t-SNE and other embedding techniques.
• Clustering: Spectral Clustering with adjustable parameters.
• Evaluation Metrics:
  • Silhouette Score
  • Adjusted Rand Index (ARI)
• Performance Analysis: Optimal cluster number selection using:
  • Elbow Method
  • Silhouette Scores
• Visualization: Interactive and static plots for embeddings and clusters.

---

🚀 Usage

Main Components:

1. Mnist_tSNE_SC.py:

   • Runs t-SNE and Spectral Clustering on the MNIST dataset.
   • Saves results in /results/Mnist.

2. Cifar_tSNE_SC.py:

   • Runs t-SNE and Spectral Clustering on the CIFAR-10 dataset.
   • Saves results in /results/Cifar-10.

3. Jupyter Notebooks:

   • Interactive exploration and visualization of results.

---

📁 Results

• Results are saved in the /results directory.
• Separate folders for MNIST and CIFAR-10 datasets.
• Includes:
  • Cluster visualizations
  • Performance metrics (Silhouette Scores, ARI, etc.)
  • Embedding plots in 2D and 3D.

---

🛠️ Requirements

• Python 3.x
• Libraries:
  • numpy
  • scipy
  • matplotlib
  • scikit-learn
  • tensorflow (for CIFAR-10)

Install dependencies using:

pip install -r requirements.txt

---

📊 Example Visualizations

t-SNE Visualization:

Spectral Clustering:

---

📜 License

This project is licensed under the MIT License.