Hi,
I am interested in decomposing an irregularly spaced time series and getting
results similiar to that obtained with the stl command for a regularly spaced
time series. I would like to know if any of the time series packages like zoo
can be used for this. From my search, I was only able to find some help with the
tseries and pastecs passages, But I had difficulty in going the full length even
with these packages . Let me explain with my sample code :
# Attemts with the functions irts, regul and tsd
n<-20
t1 <- runif(n)
n1<-(1:n)*1e6
t2<-t1*1e3
t3<-n1+t2
u <- rnorm(n)
n2<-rep(1:5,times=4); u<-u+n2
library(tseries)
x <- irts(t3, u) # could not find a decomposition method for irts
y<-data.frame(day=x$time,val=x$value)
y$day<-as.Date(y$day)
y$nday<-as.numeric(y$day)
y
with(y,plot(nday,val,type="b"))
library(pastecs)
y1<-y$day[1]
yf<-"y-m-d"
reg.y<-regul(x=as.numeric(y$day),y=y$val,units="days",methods=c("l"),
datemin=y1,dateformat=yf,deltat=5)
I get the following error message :
Error in approx(x, y, xout, method = "linear", rule = rule) :
need at least two non-NA values to interpolate
I would like to get help on the following points :
1. The actual decomposition is suposed to work with the tsd command. Is it
possible to use it without first using regul?
2. Can I succeed with the regul command by a better choice of argument values?
My attempts to set rule=2 did not help.
3. Would it be better to first get a regularly spaced time series by
interpolation, and then try decompostion with stl?
I would appreciate some practical help here.
4.Is it hopeless to attemt decomposition when the irregularity level is high?
The series that I am working on is
fairly regular in periods interpersed with either breaks or irregular data. I
would like to see the trends and seasonal effects there.
Is there an alternative method of approaching this task?
Thanking you,
Ravi
I would first abandon time-domain techniques altogether for irregularly spaced
series. The Lomb Scargle periodogram is available for R, here:
http://research.stowers-institute.org/efg/2005/LombScargle/R/index.htm.
KeithC.
-----Original Message-----
From: ravi [mailto:rv15i at yahoo.se]
Sent: Friday, February 26, 2010 3:05 PM
To: r-help at r-project.org
Subject: [R] decomposing an irregularly spaced time series
Hi,
I am interested in decomposing an irregularly spaced time series and getting
results similiar to that obtained with the stl command for a regularly spaced
time series. I would like to know if any of the time series packages like zoo
can be used for this. From my search, I was only able to find some help with the
tseries and pastecs passages, But I had difficulty in going the full length even
with these packages . Let me explain with my sample code :
# Attemts with the functions irts, regul and tsd
n<-20
t1 <- runif(n)
n1<-(1:n)*1e6
t2<-t1*1e3
t3<-n1+t2
u <- rnorm(n)
n2<-rep(1:5,times=4); u<-u+n2
library(tseries)
x <- irts(t3, u) # could not find a decomposition method for irts
y<-data.frame(day=x$time,val=x$value)
y$day<-as.Date(y$day)
y$nday<-as.numeric(y$day)
y
with(y,plot(nday,val,type="b"))
library(pastecs)
y1<-y$day[1]
yf<-"y-m-d"
reg.y<-regul(x=as.numeric(y$day),y=y$val,units="days",methods=c("l"),
datemin=y1,dateformat=yf,deltat=5)
I get the following error message :
Error in approx(x, y, xout, method = "linear", rule = rule) :
need at least two non-NA values to interpolate
I would like to get help on the following points :
1. The actual decomposition is suposed to work with the tsd command. Is it
possible to use it without first using regul?
2. Can I succeed with the regul command by a better choice of argument values?
My attempts to set rule=2 did not help.
3. Would it be better to first get a regularly spaced time series by
interpolation, and then try decompostion with stl?
I would appreciate some practical help here.
4.Is it hopeless to attemt decomposition when the irregularity level is high?
The series that I am working on is fairly regular in periods interpersed with
either breaks or irregular data. I would like to see the trends and seasonal
effects there.
Is there an alternative method of approaching this task?
Thanking you,
Ravi
I have spent quite some time building a general framework for unevenly-spaced time series. The methods do not rely on a transformation to equally-spaced data, but allow to analyze unevenly-spaced time series in their unaltered form. Please see http://www.eckner.com/research.html for my notes. In addition, I have written a paper is about trend and seasonality estimation for unevenly-spaced time series. I hope you will find the results helpful! -- View this message in context: http://r.789695.n4.nabble.com/decomposing-an-irregularly-spaced-time-series-tp1571420p4646176.html Sent from the R help mailing list archive at Nabble.com.