I have some trouble translating a model from PyMC3 about zhusuan HOT 4 CLOSED

A better name for the variable not_cheated is cheated, since a 1 means that the student cheated and a 0 means that he didn’t cheat.

from zhusuan.

Basically you specified a joint distribution (cheating_freq', 'not_cheated and observation which you forgot to include), but only sampled a subset of the variables (cheating_freq). You could

• include not_cheated into the sampler, but it's discrete, and HMC can't work on it without further augmentations;
• derive the joint distribution in which not_cheated is marginalized, but it's hard.

So I guess instead of a single HMC that samples everything at once, you should use Gibbs like the PyMC3 tutorials did.

from zhusuan.

I did try several configurations, but I kept getting python errors; it’s good to know that there was no way of succeeding with the HMC algorithm, thanks! I considered it an almighty algorithm. Perhaps I should take the Probabilistic Graphical Models specialization on Coursera.
I didn’t see the Gibbs sampler in the docs.

from zhusuan.

from zhusuan.