All tests were performed by "nn_lsi_cluster" - simple test for NN subsystem.
NN subsystem includes Radius-based Competitive Learning Neural Network
based on "rcl.v5" by Udo Heuser.
NN-Experiments Package can be downloaded here.
Last version of tested NN subsystem can be downloaded here.
To speed up clusterisation compression of input vectors was used : In this experiments NN Profiles were compressed by LSI.
After it time of clusterisation decreases from several minutes to 1-2 seconds.
Description of LSI compression is HERE
LSI subsystem can be downloaded here.
All input vectors were scaled at [-1..1]. First Model of scaling was used (maximum and minimum for all dimentions of all vectors was found and vectors were scaled all together)
NN was tested on html documents from different collections : Museums, Benchmarks, Research_Groups, Travelogues.
Here are two sets of experiments :
Questions and comments send to Dimitrij Petoukhov.