Please see here for previous work for dynamics time warping (DTW): https://github.com/ngyiuwai/PolyU-COMP5940-Final-Dissertation-Improve-UCR-Suite-by-Low-Resolution-Technique
In the final dissertation, I focus on accelerating computation of Euclidean distance only. It is because the lower bound of DTW (i.e. LB_Keogh) is a lower bound of Euclidean distance.
This is an algorithm to accelerate non-segmented sequential search for time series data with Euclidean distance as distance function. The state-of-art method is UCR Suite: https://www.cs.ucr.edu/~eamonn/UCRsuite.html.
You may integrate this algorithm with UCR Suite, because LB_Keogh in UCR Sutie is a lower bound of Euclidean distance.
- See Presentation Slides.pdf for explanation of this algorithm.
- See Demo Slidesp.pdf for visualization of this algorithm.
- See Dissertation Extract.pdf for a short summary.
The Python scripts in source code folder is a demonstration of this algorithm. Readme.txt inside the folder is a guideline for using this program. Library Matplotlib is needed if you wish to visualize the code. No external library is needed if you just want to find k-nearest neighbours.
I included some documentation in Python scripts. Hope this can help you understanding this algorithm.
Please note that the original diseertation is submited to The Hong Kong Polytechnic University in June 2020. You may find it in PolyU Library soon.