[R] Inference on parameters with package "PhaseType"

Louis Aslett aslett at stats.ox.ac.uk
Wed Sep 2 00:37:39 CEST 2015

Hi Paweł,

I replied to your direct email to me a month or so ago but perhaps your
spam filter ate it?  I'll repeat here:

Indeed, apologies the documentation could perhaps be a little clearer on
this point: the phtMCMC function is only sampling the generator matrix, not
the initial state probabilities.  My own research was motivated by problems
in reliability theory where there is no uncertainty in the initial state of
the system, so when implementing this I treated these as known (and fixed)
and only sampled from the posterior of the generator matrix entries.
Posterior sampling when the initial distribution is unknown is perfectly
possible, it's just not implemented in full generality in the package.

Adding this capability to the code should not be too big a job if this is
something you need?  Please touch base at this email address if this aspect
would be a big help and I will add to my to-do list ... would be happy to
discuss in more detail off-list.

All the best,


On 1 September 2015 at 22:29, Paweł Orliński <orlinskipawel at wp.pl> wrote:

> Hi everyone,
> I am using package "PhaseType" to approximate given sample of absorption
> times with a phase-type distribution. The function I am using is called
> "phtMCMC".
> The problem is that this function does not return parameters of starting
> state distribution. It return only parameters of subintensity matrix.
> Hence, the inference is not possible.
> Does anyone used this function and struggled with this problem?
> Or did I misunderstand something?
> Thanks in adnvance.
> Paweł Orliński

	[[alternative HTML version deleted]]

More information about the R-help mailing list