effective sample size in mcmc::chains

[Jiqiang Guo]

It seems now that when the number of chains is 1, 

this function would return the number of iterations. 

If I add the following to the end of section effective_sample_size in

stan/src/test/mcmc/chains_test.cpp

(the change is not pushed)

  stan::mcmc::chains<> c1(1, names, dimss);

  add_chain(c1, 0, "src/test/mcmc/test_csv_files/blocker1.csv", 2);

  idxs.clear();

  idxs.push_back(0);

  index = c1.get_total_param_index(c1.param_name_to_index("mu"),

                                   idxs);

  EXPECT_GT(c1.num_samples(), c1.effective_sample_size(index));

Then I would end up with

[ RUN      ] McmcChains.effective_sample_size

src/test/mcmc/chains_test.cpp:1039: Failure

Expected: (c1.num_samples()) > (c1.effective_sample_size(index)), actual: 1000 vs 1000

[  FAILED  ] McmcChains.effective_sample_size (57 ms)

--

Jiqiang

[Andrew Gelman]

That shouldn't be! Even when number of chains=1, we still split into 2 parts for split-Rhat and effective sample size, so we should get something reasonable based on these 2 chains.
A

On Jun 6, 2012, at 10:06 PM, Jiqiang Guo wrote:

> It seems now that when the number of chains is 1,
> this function would return the number of iterations.

> Jiqiang

[Jiqiang Guo]

I saw a change on the implementation.  But I am still having 

this issue when the # of chains is 1. 

--

Jiqiang