# Fast implementation of LOESS for equally spaced samples (possibly using convolution)

Is it possible to do LOESS smoothing on equally spaced data using convolution?

The Wikipedia article on LOESS mentions a possibility of using a FIR filter for LOESS. I could not find anything more about it, though.

I'm currently using statsmodels.nonparametric.smoothers_lowess.lowess in Python but it's slow. I tried switching to a Savitzky-Golay smoothing filter but I could not get comparable results on my data.

I mostly use this for detrending.

• Can you clarify your question a bit? What kind of data are you trying to smooth (or is it detrend)? Are you just looking for faster algorithm or do you want to use convolution for a specific reason? – MBaz Oct 12 '16 at 2:06
• @MBaz I need a faster implementation of lowess. The function I'm using right now is more generic and does not take into account that the data is evenly sampled. It's a biological signal. I use lowess for both detrending and smoothing since it preserves the part of the signal I care about most. The method is validated so I'm not looking for an alternative. – pkuhar Oct 14 '16 at 18:34