Modeling Missing Data

[emiruz]

HMC doesn’t like discontinuity ; so I reckon your IF statement in the signal function is part of your problem.

You can use an inverse logit to make the transition between your two densities continuous. This thread uses the trick:

Simpler Change Point Models?

One question in my mind is whether the linear nature of the interp() function will change how the change point works. I.e., the model parameters are a linear combination over the probabilities of the change point, but the discretized version of the model uses a simplex of probabilities for each discrete time point. The softmax function is highly non-linear, so it might allow for more discontinuous change breaks, whereas this version might allow for smoother change breaks as the weights evolve. O…