hafen / stlplus Goto Github PK

Seasonal-Trend Decomposition using Loess (STL) in R

License: Other

R 85.24% C++ 14.76%

stlplus's Introduction

stlplus

This package contains enhancements to the Seasonal Trend Decomposition using Loess (STL) implementation that comes with base R, stl().

Here are some of the added features over stl():

Can handle NA values
Higher order loess smoothing (more than just local constant and linear)
Automated parameter choices for local quadratic
Frequency component smoothing beyond seasonal and trend
Plot methods for diagnostics

For (very) experimental inference, prediction, and variance reduction at endpoints, see the operator package.

References

Installation

CRAN:

install.packages("stlplus")

Github:

devtools::install_github("hafen/stlplus")

License

This software is released under the BSD license. Please read the license document.

stlplus's People

Contributors

Stargazers

Watchers

Forkers

xiaosutong fsgp datavizi hadoop73 vishalbelsare qiknowledge cristianpachacama yangxhcaf elza-margolin gejielin

stlplus's Issues

forecast

I'm trying to reconstruct the original series and use it for t+1 prediction.

Anyway to use stlplus like STL with forecast package like forecast.stl ?

Trend() returning incorrect result

In writing the CV procedure I've made extensive use of the trend() method for the STL object created. Unfortunately, it seems like it produces the incorrect result. See for example:

temp<-c(-5.98,-8.15,-5.48,-2.405,-2.17,-5.985,-6.52,-5.06,-4,-6.83,-6.35,-3.63,-4.75,-7.5,-8.75,-7.42,-7.45,-8.09,-7.86,-5.91,-5.79,-8.66,-9.16,-3.575,-3.875,-7.01,-8.5,-8.07,-8.515,-10.3,-8.45,-8.67,-9.16,-9.91,-9.815,-9.12,-12.735,-14.3,-12.77,-4.66,-5.14,-7.25,-8.465,-6.525,-9.46,-11.3,-9.5,-5.86,-6.48,-8.725,-8.53,-6.73,-7.32,-7.3,-7.475,-6.575,-6.77,-7.35,-6.97,-2.85,-5.41,-6.65,-5.92,-5.08,-4.435,-6.525,NA,-5.37,-5.18,-7.575,NA,-3.7,-5.77,-9.355,-9.245,-5.98,-7.87,-9.705,-9.635,-6.795)
start<- c(1996,1); 
temp.ts <- ts(tempdat, start=start, frequency = 4)

stlobj <- stlplus(temp.ts, s.window=55, s.degree=1,
                  t.window=4,
                  t.degree=2,fc.window=10)

plot(temp.ts)
plot(stlobj)
plot(ts(trend(stlobj),start= start,
         frequency = 4))

Trend seems to return something that looks much more like the original time series than the trend line shown in the plot of the stl object. I'm sure this is just a small mistake in the method, but I'm not sure where it might be.

Interest in Implementing Cross Validation?

Hi, is there any interest in implementing cross validation for estimating missing values in the package? I have a problem where I had a time steps missing at random throughout my time series, and because of the number of time series I have to estimate the NA's for, it seemed to make the most sense to automatically perform parameter selection through cross validation.

Anyways long story short I've written up methods for K-fold and Monte Carlo cross validation to work with STLplus and I've also written a function to perform a grid search not so different than in the Scikit learn in python. If you're interested I can paste the code here, or if it's not really appropriate for the package itself I can submit my code as an answer to a self-made question on Stack Exchange.

Best,
Matt

PS: the function hasn't been super optimized yet or user-proofed, but I figured I'd include that iff it would be used by general people.

Handling NaN values

Hi,
I was searching about decompose functions that could handle NaNs and found yours, but it is implemented in R, would be possible for you to give me a brief explanation how you handle the NaNs so i can understand how it works, use it and also try to implement in python?
Thanks

Default value of t.window

Hello,
first of all I would like to thank you for creating this package. I find the diagnostic plots of the stlplus package very useful. I'm writing to you because the default value of t.window seems unusual to me. I am not sure if this is done on purpose for some reasons I don't know or if it is just an oversight.

According to the original paper "STL: A Seasonal-Trend Decomposition Procedure Based on Loess" by Cleveland et al., t.window should be set to the smallest odd integer greater or equal to(1.5*period) / (1-(1.5/s.window)(see pages 16-20 and 28). This fact is written in the documentation of the stlplus() function of the stlplus package too:

t. window: If NULL, the default, nextodd(ceiling((1.5*period) / (1-(1.5/s.window)))), is taken.

However the default value of t.windows seems to be different:

fit = stlplus(co2, s.window = 35)
fit$pars$win
  s.window t.window l.window
       35       17       13

The time series is monthly so I would expect a default t.window value equal to 19 instead of 17.

> fit = stlplus(co2, s.window = 15)
> fit$pars$win
  s.window t.window l.window
      15       19       13

And here, I would expect a t.window equal to 21 instead of 19.

Am I missing something?

Handling Missing Values

Hello,

I would like to know how the issue with Nan values has been handled within the code. By ignoring the missing data or simply by interpolation?

Regards,