# Problem in defining model, variables and parameters

So i am very new to Pyro and probabilistic programming. I just wanted to ask some thoughts whether this is feasible:

I have a set of variables v_1…v_n that are my targets (one distribution for every v_i).
Then I have some measurements: sometimes, v_i is measured itself (i know how accurately it is measured), but sometimes other parameters p_i are measured. I know the relevant equations, for example I can calculate v_i as v_i=f(p_i, v_j).

Can i use this connection to get a sense of the individual distributions for v_1 … v_n?
How would i go about this? Does anyone know a similar example that i can learn from?

My initial thoughts:

1. build a model that uses v_1…v_n; initially Gaussian distributions
2. either replace v_1 by the measurement, by using the equation for v_i if a measurement for p_i exists OR make an additional variable from it?
3. use a neural network as a guide

The biggest issue in grasping this for me is that i need v_j to calculate v_i=f(p_i, v_j). However, v_j itself could be calculated by using v_j=f(p_j, v_k) and so on. There’s some recursion issues going on. Is that an issue?

Thank you so much for taking a quick look at this post! Pointing me to examples, tutorials etc. already helps so much!