Sean M. Law
Sean M. Law
> I implemented the 8-step algo as well. I ran it on MATLAB online server, and got this: This is 8-step with `njit` and `parallel=True`? Any idea if the 8-step...
It's surprising that it jumps so dramatically at 2^24 to 2^27. I wonder if FFTW suffers from this too or if it still scales linearly. Technically, this algo should be...
> Also, the following figures confirm that six-step / eight-step FFT outperform FFTW in MATLAB Amazing! Can you confirm whether MATLAB and 6/8-step Python are single threaded or multithreaded? >...
Thanks @NimaSarajpoor. Can you please provide the raw timing data in a table? I can see that we are doing better at longer time series lengths and with 8 threads....
> If it too much, you may just take a look at the figure provided the bottom of this comment This is great! Thank you > TL; DR: It does...
@NimaSarajpoor Let's start with these comments first before we dive into DAMP.
> I think it is mostly related to the computation of sliding dot product using core.sliding_dot_product I think it is worth profiling the code to see which function most of...
> The running time of naive DAMP (stumpy-based DAMP) is ~5 sec I didn't realize that STUMPY was that performant even for `n = 100_000`! Out of curiosity, i tried...
> I am going to share a couple of observations Excellent. Please let me know if there is anything I can do to help support you!
@NimaSarajpoor Thanks for running the comparison. From what I can tell, the Python implementation seems to be around the same performance as Matlab (maybe a little faster)? What should be...