[R] vlmc - "In vlmc(traffic.clusters.stationary, cutoff = i) : alphabet with >1-letter strings; trying to abbreviate"

[Martin Maechler]

>>>>> "CA" == Constantinos Antoniou <constantinos.antoniou.rlists at gmail.com>
>>>>>     on Wed, 30 Jun 2010 12:07:16 +0300 writes:
>>>>> "CA" == Constantinos Antoniou <constantinos.antoniou.rlists at gmail.com>
>>>>>     on Wed, 30 Jun 2010 12:07:16 +0300 writes:

    CA> Dear all (copying the package author),

    CA> I have a question on the vlmc package. I am trying to
    CA> model a time series, where each element can take one of
    CA> 11 values (the result of some clustering). When I run
    CA> the following command (synthetic data to facilitate
    CA> self-contained example) 

(very good)

     > I get the following warning: ("alphabet with >1-letter strings; trying to
     > abbreviate")

     > +++ START+++
    >> library(VLMC)
    >> a <- floor(runif(1000,0,11))
    >> vc <- vlmc(a,cutoff=5)
     > Warning message:
     > In vlmc(a, cutoff = 5) :
     > alphabet with >1-letter strings; trying to abbreviate
    >> vc
     > 'vlmc' a Variable Length Markov Chain;
     > alphabet 'abcdefghijk', |alphabet| = 11, n = 1000.
     > Call: vlmc(dts = a, cutoff.prune = 5)
    -> extensions (= $size ) :
     > ord.MC   context nr.leaves     total 
     > 2        72        61      1608 
     > AIC =  5247 
    >> 
     > +++ END+++

     > The questions are: 
     > 1. What is it trying to do? 

Your contains values  0 1 2 .. 10
it tries to match them to 1-letter strings, but '10' "is 2
letters"

     > 2. How is it abbreviating? 

(not really important: using abbreviate() 

     > 3. How much should I worry about it?

not at all.  The warning is just to inform you that your input
looks a bit "unusual" to vlmc.

I do agree however, that one could argue that vlmc() should work
for inputs with values
     0:m
or   1:n
without a warning.

     > 4. What can I do?

You could use 

    vc <- vlmc(letters[1+a], cutoff=5)

to get the exact same model, but without a warning.
or
    vc <- vlmc(a, cutoff=5, quiet = TRUE)
or
    vc <- vlmc(a, cutoff=5, code1char = FALSE)



     > I have looked at the documentation plus 
     > M<c3><a4>chler M. and B<c3><bc>hlmann P. (2004) Variable Length Markov Chains:
     > Methodology, Computing, and Software.  _J. Computational and
     > Graphical Statistics_ *2*, 435-455.

That's good. 
The examples there all have a character vector (of
strings with 1 letter/character) as input.


     > Thanks for any feedback,

You're welcome!
Martin Maechler, ETH Zurich


     > -- 
     > Constantinos Antoniou, Ph.D., Assistant Professor
     > National Technical University of Athens
     > Laboratory of Transportation Engineering 
     > School of Rural and Surveying Engineering
     > 9 Heroon Politechniou st., 15780-Zografou, Athens, Greece
     > T: +30 210 7722783 - F: +30 210 7722629 
     > antoniou at central.ntua.gr - http://users.ntua.gr/antoniou


     > Dear all (copying the package author), I have a question
     > on the vlmc package. I am trying to model a time series,
     > where each element can take one of 11 values (the result
     > of some clustering). When I run the following command
     > (synthetic data to facilitate self-contained example) I
     > get the following warning: ("alphabet with >1-letter
     > strings; trying to abbreviate")



     > +++ START+++
    >> library(VLMC) a <- floor(runif(1000,0,11)) vc <-
    >> vlmc(a,cutoff=5)
     > Warning message: In vlmc(a, cutoff = 5) : alphabet with
     > >1-letter strings; trying to abbreviate
    >> vc
     > 'vlmc' a Variable Length Markov Chain; alphabet
     > 'abcdefghijk', |alphabet| = 11, n = 1000.  Call:
     > vlmc(dts = a, cutoff.prune = 5)
    -> extensions (= $size ) :
     >    ord.MC context nr.leaves total 2 72 61 1608 AIC =
     > 5247
    >> 
     > +++ END+++

     > The questions are: 1. What is it trying to do?  2. How
     > is it abbreviating?  3. How much should I worry about
     > it?  4. What can I do?

     > I have looked at the documentation plus M<c3><a4>chler
     > M. and B<c3><bc>hlmann P. (2004) Variable Length Markov
     > Chains: Methodology, Computing, and Software.
     > _J. Computational and Graphical Statistics_ *2*,
     > 435-455.


     > Thanks for any feedback, Costas



     > -- Constantinos Antoniou, Ph.D., Assistant Professor
     > National Technical University of Athens Laboratory of
     > Transportation Engineering School of Rural and Surveying
     > Engineering 9 Heroon Politechniou st., 15780-Zografou,
     > Athens, Greece T: +30 210 7722783 - F: +30 210 7722629
     > antoniou at central.ntua.gr - http://users.ntua.gr/antoniou