There's currently a bit of confusion in how times should work in discrete time. Right now, a discrete-time state evolution has a signature of x, u, t_now, t_next -> Dist. But this causes some confusion when the t values are actually discrete-time indices; for example, it seems intuitive that if I write a linear time-invariant system $x_{t+1} = A x_t + C \varepsilon_t$ and use t_observed = [0, 1, 3] that the system should be iterated twice between $t = 1$ and $t = 3$.
One proposal: change the signature of a discrete-time state evolution to x, u, k -> Dist, which seems more canonical, and is also without loss of generality (the difference $t_{k+1} - t_k$ can be included in $u_k$).
Note that this currently causes some confusion in #192, so solving it soon would be nice.
There's currently a bit of confusion in how times should work in discrete time. Right now, a discrete-time state evolution has a signature of$x_{t+1} = A x_t + C \varepsilon_t$ and use $t = 1$ and $t = 3$ .
x, u, t_now, t_next -> Dist. But this causes some confusion when thetvalues are actually discrete-time indices; for example, it seems intuitive that if I write a linear time-invariant systemt_observed = [0, 1, 3]that the system should be iterated twice betweenOne proposal: change the signature of a discrete-time state evolution to$t_{k+1} - t_k$ can be included in $u_k$ ).
x, u, k -> Dist, which seems more canonical, and is also without loss of generality (the differenceNote that this currently causes some confusion in #192, so solving it soon would be nice.