Unable to exclude warmup iterations

Summary:

Fitting a model I specified save_warmup = FALSE but on loading the stanfit CSV files they appear to still be there. Also inc_warmup = FALSE doesn't appear to make a difference in a as.data.frame.stanfit() or extract() call.

Description:

I'm using rstan version 2.18.2. I am not certain but I'm pretty sure that I didn't have this issue in the last version I was using (2.17.*).

I fit a model using the stan() function with 300 warmup iterations and 200 sampling iterations, and included the setting save_warmup = FALSE.

model_fit <- stan(
  file.path("stan_models", "general_nls_model.stan"),
  data = stan_data, 
  chains = 8,
  iter = 500,
  warmup = 300,
  control = lst(max_treedepth = as.integer(script_options$`max-treedepth` %||% 10),
                adapt_delta = as.numeric(script_options$`adapt-delta` %||% 0.8)), 
  include = TRUE,
  pars = param_to_save,
  sample_file = if (script_options$`debug-output` %||% TRUE) NULL else file.path("stanfit", str_c(output_name, ".csv")), 
  init = init_list,
  save_warmup = FALSE)

When I load the csv files (I have 8) using read_stan_csv() I get the following warning message:

Warning message:
In read_stan_csv(.) :
  the number of iterations after warmup found (500) does not match iter/warmup/thin from CSV comments (200)

Taking a look at the stanfit CSV I'm seeing the following header:

# Sample generated by Stan
# stan_version_major=2
# stan_version_minor=18
# stan_version_patch=1
# init=user
# enable_random_init=1
# seed=2062663890
# chain_id=1
# iter=500
# warmup=300
# save_warmup=0
# thin=1
# refresh=50
# stepsize=1
# stepsize_jitter=0
# adapt_engaged=1
# adapt_gamma=0.05
# adapt_delta=0.8
# adapt_kappa=0.75
# adapt_t0=10
# max_treedepth=10
# sampler_t=NUTS(diag_e)
# sample_file=stanfit/nls_model_1_1.csv
# append_samples=0

Reproducible Steps:

Here's a simpler repro of the bug:

library(rstan)

fit <- stan(model_code = "parameters { real x; } model { x ~ normal(0, 1); }", iter = 500, warmup = 300, save_warmup = FALSE, chains = 1, sample_file = "test.csv")
dim(as.data.frame(fit)) # Should only have 200 iterations
fit2 <- read_stan_csv("test_1.csv")
dim(as.data.frame(fit2)) # Should only have 200 iterations
dim(as.data.frame(fit2, inc_warmup = FALSE)) # Should only have 200 iterations
str(extract(fit2, inc_warmup = FALSE)) # Should only have 200 iterations

Current Output:

> library(rstan)
Loading required package: ggplot2
Registered S3 methods overwritten by 'ggplot2':
  method         from 
  [.quosures     rlang
  c.quosures     rlang
  print.quosures rlang
Loading required package: StanHeaders
rstan (Version 2.18.2, GitRev: 2e1f913d3ca3)
For execution on a local, multicore CPU with excess RAM we recommend calling
options(mc.cores = parallel::detectCores()).
To avoid recompilation of unchanged Stan programs, we recommend calling
rstan_options(auto_write = TRUE)
> fit <- stan(model_code = "parameters { real x; } model { x ~ normal(0, 1); }", iter = 500, warmup = 300, save_warmup = FALSE, chains = 1, sample_file = "test.csv")

SAMPLING FOR MODEL 'd4901852a59d5abb2b87f654d5e8d0a7' NOW (CHAIN 1).
Chain 1: 
Chain 1: Gradient evaluation took 8e-06 seconds
Chain 1: 1000 transitions using 10 leapfrog steps per transition would take 0.08 seconds.
Chain 1: Adjust your expectations accordingly!
Chain 1: 
Chain 1: 
Chain 1: Iteration:   1 / 500 [  0%]  (Warmup)
Chain 1: Iteration:  50 / 500 [ 10%]  (Warmup)
Chain 1: Iteration: 100 / 500 [ 20%]  (Warmup)
Chain 1: Iteration: 150 / 500 [ 30%]  (Warmup)
Chain 1: Iteration: 200 / 500 [ 40%]  (Warmup)
Chain 1: Iteration: 250 / 500 [ 50%]  (Warmup)
Chain 1: Iteration: 300 / 500 [ 60%]  (Warmup)
Chain 1: Iteration: 301 / 500 [ 60%]  (Sampling)
Chain 1: Iteration: 350 / 500 [ 70%]  (Sampling)
Chain 1: Iteration: 400 / 500 [ 80%]  (Sampling)
Chain 1: Iteration: 450 / 500 [ 90%]  (Sampling)
Chain 1: Iteration: 500 / 500 [100%]  (Sampling)
Chain 1: 
Chain 1:  Elapsed Time: 0.002715 seconds (Warm-up)
Chain 1:                0.002436 seconds (Sampling)
Chain 1:                0.005151 seconds (Total)
Chain 1: 

So far the number of iterations is correct

> dim(as.data.frame(fit))
[1] 200   2

But when loading from the CSV file, I get a warning message and the number of iterations is wrong. Aslo, inc_warmup = FALSE doesn't seem to work for as.data.frame.stanfit().

> fit2 <- read_stan_csv("test_1.csv")
Warning message:
In read_stan_csv("test_1.csv") :
  the number of iterations after warmup found (500) does not match iter/warmup/thin from CSV comments (200)
> dim(as.data.frame(fit2))
[1] 500   2
> dim(as.data.frame(fit2, inc_warmup = FALSE))
Error in .local(object, ...) : 
  formal argument "inc_warmup" matched by multiple actual arguments
> str(extract(fit2, inc_warmup = FALSE))
List of 2
 $ x   : num [1:500(1d)] -1.22 0 0 0.529 1.482 ...
  ..- attr(*, "dimnames")=List of 1
  .. ..$ iterations: NULL
 $ lp__: num [1:500(1d)] -0.745 0 0 -0.14 -1.098 ...
  ..- attr(*, "dimnames")=List of 1
  .. ..$ iterations: NULL

Expected Output:

• No warning message
• Same number of iterations (200) after loading stanfit object from CSV file
• inc_warmup option should work for as.data.frame.stanfit().

RStan Version:

2.18.2

R Version:

R version 3.6.0 (2019-04-26)

Operating System:

Distributor ID: Ubuntu
Description: Pop!_OS 19.04
Release: 19.04
Codename: disco

Start with the read_stan_csv(), as.data.frame.stanfit(), and extract() entry points using the minimal reproducible example in the issue. Compare the in-memory stan() result with the CSV-loaded result and verify that warmup rows are excluded without warnings or duplicate inc_warmup arguments; the expected result is 200 iterations in each case.