Visual Data Mining: Visalix
Visalix is a visual interface designed to facilitate man-machine cooperation on complex data analysis tasks. The Interactive Visualization paradigm allows users to better match their domain knowledge and insights with that of the processing power of a computer to analyze large datasets. Visalix proposes a number of Visual Interactive Machine Learning methods to analyze, better understand and more easily interpret the hidden structure of complex datasets. The system works in semi-supervised, unsupervised and supervised modes.
The core component of Visalix is Visual Clustering which gives a 3D interactive projection of your data and lets you manipulate it. Beyond clustering, Visalix has been designed to help users in data annotation tasks. Users can analyze a dataset, cluster items manually or by automatic optimization, annotate items, manage the label set, make label predictions for (yet) unannotated elements and visualize them, choose the most relevant item to improve current models of prediction, etc.
Visalix is domain independent. The system copes with datasets where items are described by sets of characteristics (features) and may come with their visual or textual representation. Five datasets are available in the Visalix interface for demo purposes; they have been created for different tasks in document analysis and annotation, image categorization and medical prognostics. Likewise, users can upload their datasets in standard machine learning formats.
This interface is an implementation of Loic Lecerf’s PhD thesis on Visual Interactive Machine Learning under the supervision of Boris Chidlovskii, at Xerox Research Centre Europe. The Web interface was implemented by Clément Grimal using Adobe Flash and Flex Open Source.
The Visalix Web application is available at http://visalix.xrce.xerox.com/. You can watch two demo videos or click “Run it!” to run the interface.
If someone use it, feel free to post a comment about your impressions.