This project is an interactive web-based application built using Dash, Plotly, and Bootstrap. It allows you to explore and visualize the MNIST dataset through various dimensionality reduction techniques, such as PCA, t-SNE, and UMAP. The app provides a user-friendly interface for visualizing latent spaces, individual images, and histograms of labels in the dataset.
- Dimensionality Reduction: Apply PCA, t-SNE, or UMAP to MNIST data and visualize the results in 2D or 3D.
- Interactive Plotting: Select the number of data points, latent dimensions, and reduction method with dropdowns.
- Latent Space Visualization: View the transformed space in 2D or 3D, with each point color-coded by its corresponding label.
- Image Reconstruction: Click on a point in the latent space to display the corresponding MNIST image.
- Label Distribution Histogram: View a histogram displaying the frequency of each MNIST label.
- Responsive Layout: The app is designed with Dash Bootstrap Components for a clean and responsive interface.
-
Clone the repository:
git clone [email protected]:RomainDeleat-besson/Dash_mnist.git cd Dash_mnist -
Set up the environment
- For conda users:
conda create --name mnist-visualizer python=3.10 conda activate mnist-visualizer pip install -r requirements.txt- For pip users:
python -m venv mnist-visualizer-env source mnist-visualizer-env/bin/activate pip install -r requirements.txt -
Run the app:
python MNIST_Visualizer.py -
Access the app in your web browser:
http://127.0.0.1:8050/
- Python 3.9 or higher
- Dash
- Dash Bootstrap Components
- Plotly
- numpy
- scikit-learn
- umap-learn
You can install all the dependencies via:
pip install -r requirements.txt
- Select the number of data points, reduction method (PCA, t-SNE, or UMAP), and the number of latent dimensions.
- Choose whether to display the latent space in 2D or 3D.
- Click on any point in the latent space to see the corresponding image.
- View the distribution of MNIST labels in the histogram on the right.