demo/scode.html is a 3D demo for an SCODE run in 3 dimensions, visualized with WebGL using Three.js. Training data for this demonstration is from English portion of MULTEXT-East dataset.

To create a new demo you can use the (very crude) js-data.py script. Current version only works with 13 clusters, but can easily be extended.

Usage: js-data.py SCALING_FACTOR INIT POS KMEANS UPDATES

SCALING_FACTOR is an integer that multiplied with every floating coordinate value (smaller data size)

INIT is the file containing type and its initial random vector (tab separated), i.e. the -0.702539 -0.691990 0.166098

POS is the file containing type and its gold tag (tab separated), i.e. the D

KMEANS is the file containing type and its cluster id (tab separated), i.e. the 2

UPDATES is the file containing all the updates SCODE makes until it stops, i.e. the -0.505556 -0.862696 0.012994 clocks -0.836744 0.140452 -0.529276 were -0.862178 -0.304747 0.404696 striking 0.638750 0.769366 0.008604 thirteen 0.039086 -0.605798 0.794657 . -0.450676 -0.720565 -0.526950 Winston 0.709042 0.693363 0.128475 Smith -0.003784 0.600984 0.799252 , 0.107069 -0.364476 -0.925037 his -0.144371 0.877196 -0.457912 chin 0.703353 -0.603366 -0.375824 nuzzled -0.153626 0.871067 -0.466521 into -0.775165 -0.527735 -0.347296