[Traminer-users] transition rates as substitution costs with HAM metric in seqdistmc

[Matthias Studer]

Hi Florian,

Many thanks for your bug report. The problem you have pointed out also 
affects "seqdist". Actually, seqdist and seqdistmc do not uses 
substitution costs matrix when method="HAM".

This problem will be corrected in the next version of TraMineR which 
should be released soon.

Many thanks again.
All the best.

Matthias Studer

Le 05.02.2011 14:46, Florian Hertel a écrit :
>  Hi all,
>
> first of all: thanks for TraMineR!
>
> I encountered a problem running a multiple sequence analyzes using the 
> HAM metric with substitution costs based on transition rates. I 
> started with creating a substitution cost matrix (scm) for each 
> channel based on the transition rates not allowing the rates to vary 
> over the time. The seqdistmc command (using the HAM metric) fed with 
> my scm reported, that the dimensions of the substitution matrix are 
> not equal to the number of episodes in my sequences. Thus, I created 
> an array  replicating the substitution cost matrix for each possible 
> point in time. From the report, I read that seqdistmc created a 
> "time-varying substitution cost matrix using 1 as a constant value", 
> therefore abandoning my substitution cost matrix. In order to achieve 
> my goal I just used a DHD metric based on my replicated substitution 
> cost matrix which seemed to work fine.
>
> Here a small example:
>
> dat25_1 <- matrix(c("HighCl", "LowCl", "LowCl",
>                   "LowCl", "HighCl", "LowCl",
>                   "HighCl", "LowCl", "HighCl"),
>                   3,3,dimnames=list(c("pers1","pers2","pers3"),
>                   c("class1","class2","class3")))
> dat25_2 <- matrix(c("BMW", "VW", "VW",
>                   "BMW", "BMW", "VW",
>                   "BMW", "VW", "VW"),
>                   3,3,dimnames=list(c("pers1","pers2","pers3"),
>                   c("car1","car2","car3")))
>
> seqdat1 <- seqdef(dat25_1)
> seqdat2 <- seqdef(dat25_2)
>
> subm1 <- seqsubm(seqdat1,method="TRATE",time.varying=FALSE)
> subm2 <- seqsubm(seqdat2,method="TRATE",time.varying=FALSE)
>
> subm1
> subm2
>
>
> # The first tral with error message
> seqdistmc(channels=list(seqdat1,seqdat2),method="HAM", 
> sm=list(subm1,subm2))
>
> subm13 <- array(subm1,c(2,2,3),dimnames(subm1))
> subm23 <- array(subm2,c(2,2,3),dimnames(subm2))
>
> #The second trial, ignoring my sm list but using a constant value of 1 
> for all substitutions.
> seqdistmc(channels=list(seqdat1,seqdat2),method="HAM", 
> sm=list(subm13,subm23))
>
> #The third trial, working fine (hopefully)
> seqdistmc(channels=list(seqdat1,seqdat2),method="DHD", 
> sm=list(subm13,subm23))
>
> From what I understood the first trial did not work because my 
> substitution cost matrix (scm) are only 2-dimensional. But the second 
> trial which should be "OM without indels" did not work out with my scm 
> but seqdistmc chose the default mode instead with a constant 
> substitution cost of 1. Why did the seqdist program ignored my matrix? 
> Is it because the Hamming distance is the number of positions at which 
> the corresponding states differ? If so, why is it then possible to 
> insert an own scm?
>
> Many thanks in advance and apologies for any inconvenience caused by 
> the long message.
>
> Best,
> Florian
>