These are a couple related projects which have d3.js front end interactive visualizations backed up by servers which load in data and process it on demand to send to the clients. The idea is that even if the data is large and cumbersome, lightweight visualizations can be used and efficiently delivered to the client to explore the data. Right now most of the server code uses CherryPy to write the servers so I can use fast NumPy routines for linear algebra, etc.
This work is being done in collaboration with Mauro Maggioni, Miles Crosskey, and Sam Gerber at Duke University.
The HTTP directory contains a project which involves loading in data which have been decomposed with a technique related to Geometric Wavelets. The visualizations combine an icicle view which represents the hierarchical breakdown of the data (MNIST handwritten 1s and 2s, in this case), at various scales, plus a new type of scatterplot representing the distribution of the data at that scale.
Moving further with the simplified scatterplot idea, but now for following paths in an abstract, potentially nonlinear and high-dimensional space which is represented locally by low-dimensional linear spaces.
Just tests of Autobahn and ws4py websockets implementations for passing arrays of data back and forth between client and server. No integration with the main visualizations for now.