Hello folks,

How could I do standard state-space filtering with pyMC3?

E.g. say I have a state-space model with transition equation for hidden state $x_t=g(x_{t-1}) + eta$ and observation equation $y_t=f(x_t)+eps$.

I use pyMC3 to estimate the model’s parameters using data $y_1,…,y_T$ but then how do I apply the estimated model in “online” manner?

I.e. when I get a new observation $y_{T+1}$ – how do I estimate the filtering distribution $p(x_{T+1} | y_1,…,y_{T+1})$ of the next state $x_{T+1}$? Without just rerunning the whole MCMC with an extra one datapoint added (which will be slow).

Thanks a lot in advance!