[Infovis] PEARLS Data Exploration Tool

[kamal karlapalem]

PEARLS DATA EXPLORATION

https://github.com/VijayrajS/PEARLS

https://www.youtube.com/watch?v=lh8wPMyOxVM

The PEARLS data visualization provides a conceptual visualization of high-dimensional data clusters, proposed by Mr. Nahil Jain in 2012. The algorithm aims to extract meaningful information from large quantities of data by clustering them using some already existing clustering algorithm, re-clustering the clusters to form 'pearls,' which are then assigned a particular shape and placed in the 3D space centroid of the cluster. For a more detailed description of the algorithm, refer to 'Visual Analysis of High Dimensional Real Data - Nahil Jain, 2012'<https://faculty.iiit.ac.in/~kamal/thesis_Nahil_Jain.pdf>.

The web app has the ReactJS library on the frontend and python code on the backend (hosted on the backend). Click here<https://www.youtube.com/watch?v=lh8wPMyOxVM> to see the demo video.

The PEARLS system is available for use and testing, and the code is made available for further development.
Issues and further work
·         Dealing with the file system: The system is stateless, with files in the corresponding user's folder, but the user cannot access previously uploaded files. One can add an API call to do the same and necessary additions to the frontend.
·         Vulnerabilities and security concerns: The toolkit works fine as it is in a local context. But, a web app is for server deployment, and security is the main concern (especially for valuable datasets, etc.). Basic JWT token security needs fixing of flaws in further iterations.
·         Better testing strategies for the entire toolkit: There are no proper test cases to test the app and the correctness of the backend. Implement a testing strategy (and a CI/CD pipeline-like setup if possible).
·         Graceful error handling for any corner cases found: Due to the lack of a strategy, some bugs in the system might return some generic errors. Detect such errors with more informative messages.
·         Better parallel-coordinates plot component: The frontend uses the react-vis<https://uber.github.io/react-vis/> for displaying a parallel coordinates plot for a given cluster. The framework seems to be rendering the plots but without labels. For now, a workaround is there, but a custom component for the same would be favorable.
·         Issues with the 3D plots: Users can select a pearl by using a dropdown menu. Previously, a hover-to-select pearl is with three.js. However, this was extremely slow. Identifying the bottlenecks and re-introducing this feature would be favorable.
The contacts for the project are:
            Vijayraj S                       vijayrajshanmugarajt at gmail.com
Kamal Karlapalem         kamal at iiit.ac.in
-----
Thanks and Best,
Kamal.
[Karlapalem<https://faculty.iiit.ac.in/~kamal/> :: Professor :: IIIT Hyderabad<https://www.iiit.ac.in/>]

IIIT Hyderabad :: A faculty-driven Institute for enhancing excellence among all its members!