search

andreas50 / uts

Unevenly Spaced Time Series in R
15 stars 0 forks source link

utsTrendSeasonality: trend and seasonality estimation #3

Open ironEng opened 2 years ago

ironEng commented 2 years ago

Is this package still in development? Will the trend seasonality estimation be available soon? I've read your paper on the trend and seasonality estimation methodologies for irregular time series and am particularly interested in this.

andreas50 commented 2 years ago

I recently realized that the same can be done using the "mgcv" package but with additional flexibility:

For example, if -) y_t is the quantity of interest, -) t is the the observation time expressed in the unit of a numeric year, e.g. July 1st, 2021 is 2021.5, October 1st, 2021 is 2021.75, etc., -) and frac_year = t - floor(t), then a very simple formula for gam(), e.g. y ~ s(t) + s(frac_year_pp, bs="cc"), fits a trend-seasonality decomposition. In this case, an additive decomposition with a spline for the trend and a cyclic spline for the seasonality.

My plan is to incorporate "mgcv" into the package.