Hello
I have a question regarding using old traces as starting points for new sampling runs. I am currently using the DEMetropolisZ sampler for my parameter inference. In cases where more samples are needed for convergence, I attempted to use an old trace as the starting point for a new trace. This worked, but it completely restarted the sampling process i.e. even though many samples were present, DEMetropolisZ did not use them to inform future jumps and ran as those I just initialized my solver at the final sample of the trace.
So, is it possible to input a trace with DEMetropolisZ and use the information contained to inform future metropolis jumps? Or can this not be done in PyMC3?