A first Stan program needs fixed_param=True

[seanricardoford]

The first example model, binomial-rng.stan, does not use any parameters and only has a generated_quantities block (in addition to data). Therefore, per the CmdStan manual Algorithm section, the fixed_param option needs to be set. In CmdStanPy this is done with fixed_param = True.

The quarto cell starting at line 510 should become:

N = 100
theta = 0.3
data = {'N': N, 'theta': theta}
model = csp.CmdStanModel(stan_file = '../stan/binomial-rng.stan')
sample = model.sample(data = data, seed = 123, chains = 1,
                      iter_sampling = 10, fixed_param = True,
                      show_progress = False, show_console = False)

And the following arguments section should remove the inter_warmup description and could say something like:

• fixed_param: when a model does not specify any parameters, as is the case here, this mush be True,

Importantly, let me add that this tutorial is wonderful. Thank you for taking the time to write it with CmdStanPy and Quarto.

[bob-carpenter]

Thanks, @seanricardoford. I think the problem is mainly in our doc. CmdStanPy is clever enough not to need th hint that fixed_param = True. Here's an example.

File gqonly.stan

generated quantities {
  real y = normal_rng(-0.4, 1.9);
}

>>> import cmdstanpy as csp

>>> m = csp.CmdStanModel(stan_file='gqonly.stan')
12:03:49 - cmdstanpy - INFO - compiling stan file /Users/bcarpenter/temp2/gqonly.stan to exe file /Users/bcarpenter/temp2/gqonly
12:03:53 - cmdstanpy - INFO - compiled model executable: /Users/bcarpenter/temp2/gqonly

>>> f = m.sample()
12:03:56 - cmdstanpy - INFO - CmdStan start processing
chain 1 |█████████████████████████████████████████████| 00:00 Sampling completed
chain 2 |█████████████████████████████████████████████| 00:00 Sampling completed
chain 3 |█████████████████████████████████████████████| 00:00 Sampling completed
chain 4 |█████████████████████████████████████████████| 00:00 Sampling completed
                                                                                                                                                                                                                                                                                                                                
12:03:56 - cmdstanpy - INFO - CmdStan done processing.

>> f.summary(sig_figs=2)
      Mean   MCSE  StdDev  MAD   5%  50%  95%  ESS_bulk  ESS_tail  R_hat
lp__  0.00    NaN     0.0  0.0  0.0  0.0  0.0       NaN       NaN    NaN
y    -0.42  0.029     1.9  1.8 -3.5 -0.4  2.6    4200.0    3700.0    1.0

Of course, lp__ doesn't make sense without parameters as it's just the default init of 0.

[bob-carpenter]

I'm going to leave this up rather than closing as a reminder to make this clearer to users in my doc that they don't need fixed_param = True. And I opened an issue in the docs to clarify:

stan-dev/cmdstanpy#798

[seanricardoford]

That's great, @bob-carpenter. Especially the find that CmdStanPy doesn't need fixed_param=True. Only thing to consider is that it does not like iter_warmup=0 if there are no parameters, so the code in your outstanding doc could be:

N = 100
theta = 0.3
data = {'N': N, 'theta': theta}
model = csp.CmdStanModel(stan_file = '../stan/binomial-rng.stan')
sample = model.sample(data = data, seed = 123, chains = 1,
                      iter_sampling = 10,
                      show_progress = False, show_console = False)

[bob-carpenter]

Thanks---I'll fix the doc.

This is something that could also be fixed in CmdStanPy---it probably shouldn't balk at iter_warmup=0 with fixed_param = True.