R help - Nov 2012 - depmixS4 prediction

I am getting started with using the depmixS4 package. First, I would like
to see I am very impressed with its speed and flexibility.

The question I have is regarding predicting on new data. I want to fit the
model on some sequences with observed responses, and then make predictions
on the right end of the sequences where the responses are not observed. I
see no prediction functionality anywhere, and am not sure what the best way
to formulate something like is with the package without reinventing the
wheel.

I once i have a fitted model, i would like to apply it to sequences where
the response variables on the right end of the sequence are unobserved, and
get the prediction for those (conditioned on observed covariates for the
response) using the filtering or smoothing distributions.

I could ultimately pull out the relevant parameters of the conditional
distribution of the response in each hidden state, the transition
probabilities,  rightmost posterior probability on the fully observed
sequence , and write my own code to make predictions, but am wondering if
there is a more direct way of doing it in the package.

Thanks in advance for any suggestions,

EJ

	[[alternative HTML version deleted]]

Dear EJ,

The depmixS4 package has no forecasting or predict functions, but as you
note, the forecast distribution is a relatively straightforward function of
the parameters. The posterior function provides you with the probability
distribution over the states at the end of your sequence and the transition
matrix can be used to compute the state distributions ahead.

Let me know if you have more questions, best, Ingmar

PS: you may have gotten a quicker answer had the message also been sent to
me directly.

On Thu, Nov 15, 2012 at 5:48 AM, Epic John <epicjohn@gmail.com> wrote:
> I am getting started with using the depmixS4 package. First, I would like
> to see I am very impressed with its speed and flexibility.
>
> The question I have is regarding predicting on new data. I want to fit the
> model on some sequences with observed responses, and then make predictions
> on the right end of the sequences where the responses are not observed. I
> see no prediction functionality anywhere, and am not sure what the best way
> to formulate something like is with the package without reinventing the
> wheel.
>
> I once i have a fitted model, i would like to apply it to sequences where
> the response variables on the right end of the sequence are unobserved, and
> get the prediction for those (conditioned on observed covariates for the
> response) using the filtering or smoothing distributions.
>
> I could ultimately pull out the relevant parameters of the conditional
> distribution of the response in each hidden state, the transition
> probabilities,  rightmost posterior probability on the fully observed
> sequence , and write my own code to make predictions, but am wondering if
> there is a more direct way of doing it in the package.
>
> Thanks in advance for any suggestions,
>
> EJ
>
>         [[alternative HTML version deleted]]
>
> ______________________________________________
> R-help@r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide
> http://www.R-project.org/posting-guide.html
> and provide commented, minimal, self-contained, reproducible code.
>
	[[alternative HTML version deleted]]