Non-Linear Dimensionality Reduction Visualization Tool.

polarisR

polarisR is a shiny application for diagnosing $2\text{-}D$ non-linear dimensionality reduction (NLDR) layouts generated using methods such as t-SNE and UMAP with different hyper-parameters. It also facilitates comparison of how the data appear in high dimensions through various tour methods, including scatter, sage, and slice. The application provides an interactive platform to explore high-dimensional datasets, diagnose NLDR layouts using the quollr package, and compare different NLDR results.

Features

• Data Loading: Load your own CSV datasets or use one of the example datasets provided.
• NLDR Methods: Choose between t-SNE and UMAP for dimensionality reduction.
• Interactive Visualization: Explore the 2D NLDR layout of your data with interactive plots.
• Dynamic Tour: Take a dynamic tour of the high-dimensional data to understand its structure.
• Diagnostics of NLDR Layout: Use the quollr package to diagnose the NLDR layout.
• NLDR Methods Comparison: Compare the results of different NLDR methods and hyperparameter settings.
• Side-by-Side Visualization: View two visualizations (NLDR layout and tour view) side-by-side with linked brushing for a direct comparison.

Installation

You can install polarisR from GitHub using the following commands in R:

# Install pak if you haven't already
# install.packages("pak")

pak::pak("Divendra2006/polarisR")

Usage

To run the polarisR shiny app, use the following command in R:

polarisR::run_nldr_viz()

You can also upload your own dataset in CSV format.

About the name

projective output layouts and reduced interactive surfaces in R

License

This project is licensed under the MIT License - see the LICENSE.md file for details.