CmdStan 2.7.0

[Daniel Lee]

It's been tagged.

Daniel

[Daniel Lee]

I can hold off on announcing it until we hear from at least RStan and PyStan.

Daniel

[Bob Carpenter]

There's no absolute need to keep in synch, but all else
being equal, I think it's easier on the users to put them
all out at once so nobody has to keep coming back to see when
their version is available.

- Bob

> --
> You received this message because you are subscribed to the Google Groups "stan development mailing list" group.
> To unsubscribe from this group and stop receiving emails from it, send an email to stan-dev+u...@googlegroups.com.
> For more options, visit https://groups.google.com/d/optout.

[Allen B. Riddell]

re: PyStan, I've started the branch and will have some time later this
week.

[Daniel Lee]

Thanks!

[Ben Goodrich]

On Wednesday, July 15, 2015 at 1:19:31 AM UTC-4, Daniel Lee wrote:

I can hold off on announcing it until we hear from at least RStan and PyStan.

RStan has to wait a few days for a new to StanHeaders to get accepted on CRAN, which means StanHeaders needs to get released.

[Rob J. Goedman]

Hi Bob,

Running Pkg.test(“Stan”) using CmdStan 2.7.0 works fine with the current version of Stan.jl (v”0.2.1”) on Julia-0.3.10.

Early August I will create an update for Stan.jl that works on Julia-0.4-dev. Julia-0.4 requires a couple of changes on how multiple chains are scheduled in parallel.

Regards,

Rob

Notes:

=====

Fyi, I compiled CmdStan on OS X 10.11 Beta, XCode-7.0-beta and with clang++ and O/O_STANC=3. I typically disable the template-depth section for clang in the Mac specific make. Attached the detailed output (just for the binormal example, all other tests ran fine as well).

I haven’t had time to look at the new (experimental?) features (ADVI) yet, will take that along in August as well unless someone pokes me earlier. If ADVI creates a .csv file that should be pretty simple.

--- Translating Stan model to C++ code ---

bin/stanc /Users/rob/.julia/v0.3/Stan/Examples/Binormal/tmp/binormal.stan --o=/Users/rob/.julia/v0.3/Stan/Examples/Binormal/tmp/binormal.hpp

Model name=binormal_model

Input file=/Users/rob/.julia/v0.3/Stan/Examples/Binormal/tmp/binormal.stan

Output file=/Users/rob/.julia/v0.3/Stan/Examples/Binormal/tmp/binormal.hpp

--- Linking C++ model ---

clang++ -DBOOST_RESULT_OF_USE_TR1 -DBOOST_NO_DECLTYPE -DBOOST_DISABLE_ASSERTS -I src -I stan/src -isystem stan/lib/stan_math_2.7.0 -isystem stan/lib/eigen_3.2.4 -isystem stan/lib/boost_1.58.0 -Wall -pipe -DEIGEN_NO_DEBUG -Wno-unused-function -Wno-tautological-compare -Wno-c++11-long-long    -O3 -o /Users/rob/.julia/v0.3/Stan/Examples/Binormal/tmp/binormal src/cmdstan/main.cpp -include /Users/rob/.julia/v0.3/Stan/Examples/Binormal/tmp/binormal.hpp 

Inference for Stan model: binormal_model

4 chains: each with iter=(1000,1000,1000,1000); warmup=(0,0,0,0); thin=(1,1,1,1); 4000 iterations saved.

Warmup took (0.016, 0.016, 0.016, 0.017) seconds, 0.066 seconds total

Sampling took (0.023, 0.026, 0.026, 0.027) seconds, 0.10 seconds total

                    Mean     MCSE  StdDev    5%       50%       95%  N_Eff  N_Eff/s    R_hat

lp__            -1.0e+00  3.1e-02     1.0  -3.1  -6.7e-01  -5.2e-02   1054    10393  1.0e+00

accept_stat__    9.1e-01  1.9e-03    0.12  0.66   9.5e-01   1.0e+00   4000    39443  1.0e+00

stepsize__       9.4e-01  9.9e-02    0.14  0.82   8.8e-01   1.2e+00    2.0       20  9.2e+13

treedepth__      1.9e+00  6.6e-02    0.54   1.0   2.0e+00   3.0e+00     67      663  1.0e+00

n_leapfrog__     3.1e+00  2.6e-01     1.6   1.0   3.0e+00   7.0e+00     39      382  1.0e+00

n_divergent__    0.0e+00  0.0e+00    0.00  0.00   0.0e+00   0.0e+00   4000    39443      nan

y[1]            -1.5e-02  2.1e-02     1.0  -1.7   9.4e-03   1.7e+00   2411    23772  1.0e+00

y[2]             1.9e-02  1.9e-02    0.98  -1.6   3.7e-03   1.6e+00   2585    25486  1.0e+00

Samples were drawn using hmc with nuts.

For each parameter, N_Eff is a crude measure of effective sample size,

and R_hat is the potential scale reduction factor on split chains (at 

convergence, R_hat=1).

Iterations = 1:1000

Thinning interval = 1

Chains = 1,2,3,4

Samples per chain = 1000

Empirical Posterior Estimates:

         Mean         SD      Naive SE      MCSE       ESS   

lp__ -1.008360828 1.0150792 0.016049811 0.030513647 1106.6533

 y.1 -0.014683177 1.0253441 0.016212114 0.022366559 2101.5559

 y.2  0.019096266 0.9790049 0.015479427 0.016814434 3390.0431

Quantiles:

        2.5%       25.0%       50.0%       75.0%        97.5%   

lp__ -3.6776187 -1.4042550 -0.670830500 -0.29177000 -0.024094995

 y.1 -2.0134163 -0.7226885  0.008652330  0.65454450  2.071252500

 y.2 -1.8838548 -0.6644820  0.003470805  0.67720425  1.967234250

                   PSRF           97.5%     

         lp__   1.004000×10⁰   1.0090000×10⁰

accept_stat__   1.140000×10⁰   1.3140000×10⁰

   stepsize__ 3.0214979×10¹⁴ 5.89831932×10¹⁴

  treedepth__   1.029000×10⁰   1.0870000×10⁰

 n_leapfrog__   1.036000×10⁰   1.1040000×10⁰

n_divergent__            NaN             NaN

          y.1   1.000000×10⁰   1.0010000×10⁰

          y.2   1.001000×10⁰   1.0020000×10⁰

 Multivariate            NaN             NaN

     Z-score p-value

lp__   0.158  0.8746

 y.1  -0.321  0.7486

 y.2   0.610  0.5416

     Z-score p-value

lp__  -0.777  0.4370

 y.1  -0.404  0.6865

 y.2  -1.166  0.2438

     Z-score p-value

lp__   0.339  0.7350

 y.1   0.492  0.6226

 y.2   0.482  0.6298

     Z-score p-value

lp__   1.424  0.1545

 y.1   1.014  0.3104

 y.2   0.354  0.7232

     95% Lower  95% Upper 

lp__  -3.09777 -0.00034912

 y.1  -2.01522  2.07113000

 y.2  -1.78774  2.04616000

         lp__          y.1          y.2    

lp__  1.000000000 -0.020537582 -0.040924916

 y.1 -0.020537582  1.000000000  0.066567001

 y.2 -0.040924916  0.066567001  1.000000000

        Lag 1         Lag 5         Lag 10        Lag 50   

lp__  0.44395293    0.028651176  -0.0053760634 -0.006793620

 y.1  0.30353776    0.018117719   0.0043199533 -0.067077180

 y.2  0.25876139    0.027311501  -0.0089279184  0.024112327

        Lag 1         Lag 5         Lag 10        Lag 50   

lp__ 0.540196036    0.085138378  0.04371068130  0.032327092

 y.1 0.241527920    0.009184393 -0.00038747966 -0.041938622

 y.2 0.071547072   -0.027107125  0.03388795739 -0.012981663

        Lag 1         Lag 5         Lag 10        Lag 50   

lp__  0.46308768   -0.010879561    0.094918605 -0.031567440

 y.1  0.28269055   -0.009826277    0.012703330 -0.002452647

 y.2  0.24493764   -0.019832863   -0.034029837 -0.042601331

        Lag 1         Lag 5         Lag 10        Lag 50   

lp__   0.6145697  0.13338730371    0.101844429 -0.026271114

 y.1   0.2426259 -0.00988255455   -0.014203974  0.036094890

 y.2   0.2848844  0.00097344396   -0.014377092 -0.015794737

[Bob Carpenter]

Great. We're trying to keep the external interfaces for CmdStan
stable.

There's now variational inference if you want to figure out how to
add that, but it's still "experimental" --- we haven't evaluated on
lots of models.

- Bob

> --
> You received this message because you are subscribed to the Google Groups "stan development mailing list" group.
> To unsubscribe from this group and stop receiving emails from it, send an email to stan-dev+u...@googlegroups.com.
> For more options, visit https://groups.google.com/d/optout.

> <binormal-summaryplot-1.pdf>