Hi,
Is there a way to get the convergence check informations programatically ?
I mean having these information (number of diverging samples, rhat, etc…) in variables in order to use them in scripts …
Thanks for your help.
H.
Hi,
Is there a way to get the convergence check informations programatically ?
Thanks for your help.
H.
Yup - you can extract the \hat{R} values using pm.summary
while the presence of divergences can be extracted directly from the trace. Here’s an example:
with pm.Model() as model:
x = pm.Normal('x')
trace = pm.sample()
n_diverging = trace['diverging'].sum()
rhat = pm.summary(trace)['r_hat']
To obtain other statistics from the sampler, you can see the variables listed via trace.stat_names
.