I have a time series "A" and another one "B". I would like to find occurrences of "B" inside "A". Typically, "A" is much bigger (magnitude: millions of points) than "B" (magnitude: hundreds of points)
- I don't necessarily want exact matching (obviously, since the values we deal are real numbers not, say, text) The values are preferably close, though.
- Trends/shapes are more important than exact time matching (i.e. somehow Dynamic Time Warping could be useful)
- With regards to the first two points, what happens if I don't care about the absolute values at all but just the shape (maybe taking a log of the values?)
- What happens if I would like to scale the number of points to billions from millions? How to scale the system? How do I index a time series? (maybe use Mahout, Lucene, somehow?)
What's inside above parentheses are the results of my short research, not necessarilly correct. All in all, what kind of an approach should I take? What are your general suggestions/tips? I'd like to develop a system, so any practical examples/suggestions are even more welcome.
Thanks!