It looks like it's significantly faster (13 milliseconds) if using
scipy.signal.convolve - even though that adds another dependency to the
application, I'm considering allowing it this time.
--
Rune / 5Q5R
toggle quoted message
Show quoted text
On Fri, 25 Oct 2019 at 11:17, Oristo <ormpoa@...> wrote:
I don't know what "convolve" does,
but I assume it's something quite complicated?
"convolve" is a rolling sum.
For this special case of convolving with constant one,
it can be speeded by dropping the oldest value and adding the newest to
the sum
for all ones in e.g. self.step
This is AKA 1-D box filter or summed area table
