{cmdstanr} variational() ignores eta parameter

Interfaces
,
1

I’m using the variational() function and setting eta = 1, but that doesn’t seem to matter: Stan is still trying to find eta. Sorry if I’m misunderstanding this option (new to variational Bayes).

some_data |>
  cell_line_model$variational(eta = 1)

Give me

...
Gradient evaluation took 0.248526 seconds 
1000 transitions using 10 leapfrog steps per transition would take 2485.26 seconds. 
Adjust your expectations accordingly! 
Begin eta adaptation. 
Iteration:   1 / 250 [  0%]  (Adaptation) 
Iteration:  50 / 250 [ 20%]  (Adaptation) 
Iteration: 100 / 250 [ 40%]  (Adaptation) 
Iteration: 150 / 250 [ 60%]  (Adaptation) 
Iteration: 200 / 250 [ 80%]  (Adaptation) 
Success! Found best value [eta = 1] earlier than expected. 
...

Operating System: MacOS 14.4.1
Interface Version: 0.7.1 (cmdstan v2.34.1)
Compiler/Toolkit:

Thanks,
Karim

2

I believe you need to pass adapt engaged=0 (which seems to be adapt_engaged=FALSE in R), otherwise the user-provided value is just used as a starting point for adaptation, not the final value

3

Thanks!