Welcome to robust-ts: Robust Time Series project!

"robust-ts" is a collaborative project to provide robustifications to the basic time series procedures from package stats. A target will be chapter 8 in "Robust Statistics, Theory and Methods" by Maronna, Martin and Yohai; 2006.

Guidelines for package robust-ts: 

naming convention 

• yet to be announced (to be decided by plenary vote in r-sig-robust)
  
• open issues
• package & function names: suffix vs. prefix ``rob'' / ``robust''
• captialization, separators
  
• part ts is agreed upon
  

programming in layers/stages

• first layer/stage (implementation of algorithms)
  

• implementation of the algorithm in R
• if necessary: delegation of critical code to C / C++ or FORTRAN
• basic interfacing functions (with large number of arguments)
  
• no particular data types (apart from vectors and  matrices)
• common "style" not necessary
  
• no encapsulation
• gives "alpha version" --- extensive testing at this stage possible and necessary
  

• second layer/stage (user interface)

• for consistency within this package: common "style" necessary
• use of particular classes for input and output
  
• low number of structured arguments in user-interface functions/methods
  
• mimic existing user interfaces
• use of data_frames in regression type setup
  
• use of formula type model definitions
• accessor and replacment functions for the slots of class
  
• particular methods
  
• print
  
• show
• plot
• summary
  

input structure

• possible input data:
  
• (raw) vectors,
  
• ts objects
• zoo objects (if method allows for irregularly spaced time stamps)
  
• function argument list / signature:
  as far as possible same arg's as classical routine
• additional arguments split into control/``design type''/method type arguments:
  • ``design'':  tuning parameters for the robust procedure which are
    
  • subject to frequent change by the standard user;
    
  • explicit and transparent to the user
    
  • e.g. window width in some filters
    
  • control: tuning parameters for the robust procedure  
  • for which the standard user will use default values
    
  • these are packed into control object
  • may be generated by a function control --see robustbase
  • the author will provide control-functions for each of his provided methods (borrow from robustbase)
  • method argument: used in a ``super''-function to determine the actual robustification / procedure
    two different types of values should be allowed simultaneously
    • (a vector of) characters:
      
    • for: default/standard user
    • names of the robustification(s)/ procedure(s)
    • vector-valued: several methods may be calculated in parallel / in one loop over the time stamps
    • an object of a S4-method-class providing the functions to be used for this/these actual procedure(s)
      
    • for: ``power user''
      
    • goal: a ``power user'' should be able to define his own method-class
      to make the routine use the user's code without due intervention of the author
      
    • usually this object will be generated by a generating function
      
    • the author will provide a couple of method-classes and generating functions

output structure

• time-stamped elements:
  
• at least of type  ts,
  
• if irregular input zoo
• output type
  
• at least: S3-classes,
  
• better: S4-classes
• inheritance
• class should ``inherit'' from corresponding classical return class
  
• if return object of robust procedure is S4 and of the classical one is S3:
  S4-class should have the same slots as classical S3-class
• in case of several methods computed in parallel:
  
• list of corresponding objects ->
  
• particular plot/print methods for such lists
  
• particular consistency checks / validity methods
  
• particular / overloaded methods like print.acf or getMethod("print","acfrob")
  • if possible, every argument legal for least specific method
    (eg. plot.default) should be usable in particular (plot-)function

Links to additional documents

• Open issues as to using R-Forge
  
  
• "HOWTO": What you have to do to collaborate in 10 steps

Maintained Target / Todo-List for Project "robust-ts"

NOTE:

a sort of legend

• column "source of code" is to capture the fact that maybe we get code from outside,
  in particular from Insightful...
• column "modus" will mention if original code resides in another R package and is
  simply integrated by some wrapper --- as in the [g]arch case
• more than one person may be listed in the "working on it"; the "from" and "to"
  columns are meant to show in which time period a certain collaborator has been/ was
  working on this topic
  

The project summary page you can find here.