9 ms·
How is this so fast? I thought to compare charts like these you have to slide every chart over every offset to get the best match?
by notintokyo 15y ago
How is this so fast? I thought to compare charts like these you have to slide every chart over every offset to get the best match?
- waffle_ss 15y agoThey detail their algorithm in the "Methods" section at the end of their whitepaper: http://www.google.com/trends/correlate/whitepaper.pdf http://www.google.com/trends/correlate/whitepaper.pdf
- danvk 15y ago(I work on correlate) The gist of why this is so fast is: 1. For the vast majority of queries, we only need to compare hashes of the time series, not the time series themselves. 2. We make no attempt to detect shifted correlations. The correlations you see are all synchronized in time.