Elastic Product Quantization for Time Series

Analyzing numerous or long time series is difficult in practice due to the high storage costs and computational requirements. Therefore, techniques have been proposed generate compact similarity-preserving representations of series, enabling real-time similarity search on large in-memory data collections. However, existing are not ideally suited for assessing when sequences locally out phase. In this paper, we propose use product quantization efficient similarity-based comparison under warping. The idea first compress by partitioning into equal length sub-sequences which represented a short code. distance between two can then be efficiently approximated pre-computed elastic distances their codes. forces unwanted alignments, address with pre-alignment step using maximal overlap discrete wavelet transform (MODWT). To demonstrate efficiency accuracy our method, perform an extensive experimental evaluation benchmark datasets nearest neighbors classification clustering applications. Overall, solution emerges as highly (both terms memory usage computation time) replacement measures