COSIATEC and SIATECCompress:Pattern discovery by geometric compression

Three versions of each of two greedy compression algorithms, COSIATEC and SIATECCompress, were run on the JKU Patterns Development Database. Each algorithm takes a point-set representation of a piece of music as input and computes a compressed encoding of the piece in the form of a union of translational equivalence classes of maximal translatable patterns. COSIATEC iteratively uses the SIATEC algorithm to strictly partition the input set into the covered sets of a set of MTP TECs. On each iteration, COSIATEC finds the “best” TEC and then removes its covered set from the input dataset. SIATECCompress runs SIATEC just once to get a list of MTP TECs and then selects a subset of the “best” TECs that is sufficient to cover the input dataset. Both algorithms select TECs primarily on the basis of compression ratio and compactness