|laziness, impatience, and hubris|
Re: Counter - number of tags per intervalby LanX (Abbot)
|on May 04, 2013 at 22:08 UTC||Need Help??|
Just another idea based on tuning your current algorithm which is quite clever! (Well after being understood - you know my opinion about clear naming ;-)
After analyzing the numbers you provided, I suppose that intersections of different ranges are comparatively rare and if they appear they stretch over long distances.
So it doesn't really make sense to go into the count loop for each individual point, just go on counting the "1" till you reach the next "boundary" and add them collectively to your counter instead incrementing them one by one with++.
The boundaries are nothing else as the sorted list of range starts-1 and stops. In your example for ID "b"
this means [9,11,15].
and results in the following flow
As you see, you can also stop counting between ranges and even try to jump forward in the file with seek.
3e6 Intervals/50 IDs *2 Borders = 120000 Borders
6e9 Points / ID
=> in average 6e9/12e4=50000 points between two adjacent borders, where you can avoid the whole count loop.
This should lead to a considerable boost!
( addicted to the Perl Programming Language)