Skip to content

Best datatype for time-series with irregular ocurrencies

4 messages · jmoreira@fe.up.pt, Gabor Grothendieck, Diethelm Wuertz +1 more

Original
jmoreira@fe.up.pt
# 
Hello,

Does someone has experience working with time-series where the ocurrencies do 
not happen at regular time spaces? In one day I can have 20 ocurrencies and in 
the following day I can have the double. ts datatype must have regular time 
spaces. What is the best way to put the data if I want to use forecasting 
methods as Holt-Winters, NN or SVM? To put data and time in different 
variables? If in one variable it would be of Date datatype?

Thank you for any help.

Joao


-------------------------------------------------
Joao Mendes Moreira
Faculdade de Engenharia da Universidade do Porto
DEMEGI / GEIN
Gabor Grothendieck
# 
<jmoreira <at> fe.up.pt> writes:

: Does someone has experience working with time-series where the ocurrencies 
do 
: not happen at regular time spaces? In one day I can have 20 ocurrencies and 
in 
: the following day I can have the double. ts datatype must have regular time 
: spaces. What is the best way to put the data if I want to use forecasting 
: methods as Holt-Winters, NN or SVM? To put data and time in different 
: variables? If in one variable it would be of Date datatype?

There are several packages that can accommodate irregularly spaced time series.
Regarding your specific requirements, zoo is the only one that supports 
the Date class as the time variable.  Also, the upcoming version of zoo 
(not yet on CRAN) has support for e1071::svm and nnet::nnet .

There is a summary of the various irregular time series classes/packages
here:

https://stat.ethz.ch/pipermail/r-sig-finance/2004q4/000210.html
Diethelm Wuertz
#
jmoreira at fe.up.pt wrote:

            

      
      
      
    

        
package fBasics from Rmetrics (www.rmetrics.org)  has S4 timeDate and 
timeSeries objects similar
to those in SPlus for irregular time series manipulations and 
investigations.

package its is another option

Regards Diethelm Wuertz

            

      
      
      
    

        

      
      
    

  
  


  


      

    

    

      
      

        


  

        

      Achim Zeileis
    

    

      
      #
    

  


  

      
On Wed, 15 Dec 2004, Diethelm Wuertz wrote:

        

            

      
      
      
    

        
Just to be complete: in addition to
  "its" in package its
  "timeSeries" in fBasics
which are S4 classes, there are
  "irts" in package tseries
  "zoo" in package zoo
which are S3 classes for irregularly spaced observations. "its" is
probably the most mature, "timeSeries" is - as Diethelm said - similar to
the S-PLUS implementation and "zoo" has the advantage that time
information can be of (almost) arbitrary class.
The development version of zoo is due for release next week or so,
contact me off-list if you want the current version.

Best,
Z