Resetting model variables to sample for new data

Can/Should we reset the model distribution parameters to its original prior values before repeatedly setting new observations and sampling?

For instance, assume the following pseudo code:

with pm.Model() as mm:
    p1 = pm.somedistrib(param=hyper)
    p2 = pm.anotherdistrib(param1=p1, param2=hyper)

    pm.Potential("", (p1,p2, observed1))
    trace1 = pm.Sample(...)

    ***# when we call here i would like to use the default/original values of p1,p2 - not any posterior updates***
    pm.Potential("", (p1,p2,observed2))
    trace2 = pm.Sample(...)

Is it even required to reset p1,p2 in this flow if i want to use the original prior values, or calling Potential/Sample would not change any state inside p1,p2 and all changes/value of the parameters will be collected inside the trace?


You don’t need to reset anything. The same priors will be used when you call sample again