Open neukym opened 2 months ago
Top is downsample at 5 milliseconds, bottom is every 2ms. While 2ms clearly reveals another spike at 35ms (which is very good), it does so at the cost of messier STL (for the moment, this may change if we improve preprocessing).
The downsampling adds a speed cost of t*log(t).
Going up to 9 derangements improves this:
So maybe it should be 2ms and 9 derangements.
Top is downsample at 5 milliseconds, bottom is every 2ms. While 2ms clearly reveals another spike at 35ms (which is very good), it does so at the cost of messier STL (for the moment, this may change if we improve preprocessing).
The downsampling adds a speed cost of t*log(t).