RAP: a computer program for exploring similarities in behavior sequences using random projections

Abstract

A computer program (RAP, for random projection) for exploring similarities between and within sequences of behavior is presented. Given a time window of a sequence, the program calculates a signature, a real-valued vector that is a random projection of the contents of the window (i.e., the codes occurring within it and their relative location, or onset and offset times) into an arbitrary K-dimensional space. Then, given two different time windows from the same sequence or from different sequences, their similarity is computed as an inverse function of the Euclidean distance between their respective signatures. By defining moving (overlapped or not overlapped) windows along each sequence and calculating similarities between every pair of windows from the two sequences, a map of similarities or possible recurrent patterns is obtained; the RAP program represents them as gray-level lattices, which are displayed as mouse-sensitive images in an HTML file. Computation of similarities is based on the random projection method, as presented by Mannila and Seppänen (2001), for the analysis of sequences of events. The program reads sequence data files in Sequential Data Interchange Standard (SDIS) format (Bakeman Quera, 1992,1995a).