estimate parameters of state space model

[Sergio]

Hi all,

I'm quite new to tf probability and I've recently started to use the sts library.

How ca I estimate the latent state noise and the observation noise using both, mcmc and variational inference?

As an example, let's say that I have a time series that follows a random walk:

model = sts.LocalLevel()

model = sts.Sum([model])

sts_model = model.make_state_space_model(num_timesteps=100, param_vals=[observation_std, state_std])

sample = sts_model.samlpe(1)

Now my sample is drawn from the state space model distribution. Let's say that I don't know the  observation_std and state_std, how can I use mcmc and vi to estimate them from the sample? Could somebody provide an example?

I've already posted the question as a github issue because I didn't know this group existed, but I believe this is a better place for such a question.

Any help is very appreciated!

Sergio

[Dave Moore]

There are examples of fitting STS model parameters using VI in the example notebook:

https://github.com/tensorflow/probability/blob/master/tensorflow_probability/examples/jupyter_notebooks/Structural_Time_Series_Modeling_Case_Studies_Atmospheric_CO2_and_Electricity_Demand.ipynb

For MCMC, you might want to check out `tfp.sts.fit_with_hmc` (https://www.tensorflow.org/probability/api_docs/python/tfp/sts/fit_with_hmc), which does a lot of the work for you.

Dave

--
You received this message because you are subscribed to the Google Groups "TensorFlow Probability" group.
To unsubscribe from this group and stop receiving emails from it, send an email to tfprobabilit...@tensorflow.org.
To view this discussion on the web visit https://groups.google.com/a/tensorflow.org/d/msgid/tfprobability/bf0863b6-12b3-4332-811e-d0c0670e4a7an%40tensorflow.org.