The documentation says
- vars ( list ) – List of variables for sampler
but is the intended effect when sampling? Or more bluntly put what’s the point of the argument and what’s it used for?
https://docs.pymc.io/api/inference.html#module-pymc3.step_methods.metropolis
P.S. I’m going for a really screwy inference run on purpose 
Example
That’s for specifying the variable you want the sampler to sample. After you initialize a model, you can do model.free_vars which gives you a list of free parameters in your model. And then you can specify different sampler to sample different vars. For example see: https://docs.pymc.io/notebooks/sampling_compound_step.html
Thanks jungpenlao,
In the notebook then I specified vars=(mu,) but the sample seems to still be sampling the distribution of sd? My understanding is that pymc3 would only be sampling for value of mu but It looks like I’m mistaken
usually we passed vars=[mu] to it (not sure if it makes any differences tho)
Made the change but still confused. Also read the doc you sent but too dense to figure out what it means in reference to my confusion 
With vars specified still looks like ‘sd’ is still sampled. If I understand correctly only mu should be sampled? Or does the step argument in my case mean that ‘mu’ is sampled first?
Below is the graph from the notebook where I’m seeing 2 chains for both mu and sd leading to my confusion
Thank you for the help
Oh I see your confusion, all free parameters will be sampled or approximated, the var arg is to specify which sampler (eg metropolis) sample what (eg mu), otherwise it will be assigned automatically.
Ah ok that makes sense, So in this cause I could have mu sampled by MetropolisHastings and sd sampled by NUTS, for example
Exactly.
Thank you!
