Solved: Re: Trouble with ODS Output PostSumInt

mcraft · Posted 12-01-2021 01:54 PM

Hello,

I'm trying to fit a model using PROC MCMC and output credible intervals for each parameter. I add the option ods output PostSumInt = intervals, which should output by default. Below is the code I've specified for fitting the model.

ods output Corr = mycorr
Cov = mycov
DIC = mydic
PostSummaries = myparams
Geweke = mygeweke
PostSumInt = intervals; * output ods tables of interest to datasets;
PROC MCMC
data = data SEED=9879 NMC=100000 NBI=500 MAXTUNE=50 THIN=25 STATISTICS = summary DIAGNOSTICS = GEWEKE STATS = CORR STATS = COV DIC;

ARRAY location_scale_mean[2] (0,0);
ARRAY location_scale_cov[2,2] (1,0,0,1);
ARRAY location_scale_rf[2] sub_1 sub_2;

PARMS b0=0.02 b1=0.22 b2=-0.07 b3=0.21 b4=0.13
tau0=0.17 tau1=0.11 tau2=0.13 tau3=0.21 tau4=-0.15
lp0=0.25 lp1=0.12 lp2=-1.08
cov=-0.09 ln_varScale=0.33;

BY rep_n;
WHERE rep_n < 4;

PRIOR b: ~NORMAL (0, VAR=1000);
PRIOR tau: ~NORMAL (0, VAR=1000);
PRIOR lp: ~NORMAL (0, VAR=1000);
PRIOR cov: ~NORMAL (0, VAR=1000);
PRIOR ln_varScale~NORMAL(0,VAR=1000);

varBS = EXP(lp0 + W1_c*lp1 + W2_b*lp2);
varScale = EXP(ln_varScale);

L11=SQRT(varBS);
L21=cov/L11;
L22=SQRT(varScale-L21**2);

mu = b0 + b1*X1_c + b2*X2_b + b3*W1_c + b4*W2_b + L11*sub_1;
varWS = EXP(tau0 + X1_c*tau1 + X2_b*tau2 + W1_c*tau3 + W2_b*tau4 + L21*sub_1 + L22*sub_2);

RANDOM location_scale_rf ~ mvn(location_scale_mean,location_scale_cov) SUBJECT=subj;
MODEL y ~ NORMAL(mu,var = varWS);

RUN;
ods output close;

After running this model, I get an error message that says:

WARNING: Output 'PostSumInt' was not created. Make sure that the output object name, label, or path is spelled correctly. Also,

verify that the appropriate procedure options are used to produce the requested output object. For example, verify that

the NOPRINT option is not used.

Any ideas on why this output was not created? I'm pretty sure it's spelled correctly and I can't find any required options for producing this output.

Thanks!

Madeline

Reeza · Posted 12-01-2021 03:03 PM

You're using an old version of SAS/STAT, the current version is 15, you're on 12.3.
The 12.3 documentation doesn't have a PostSumInt table listed, would any of the other options get you there?
https://support.sas.com/documentation/cdl/en/statug/66103/HTML/default/viewer.htm#statug_mcmc_detail...

It looks like this isn't a feature in your current version 😞

View solution in original post

Reeza · Posted 12-01-2021 02:03 PM

1. Can you please post the full log?

2. What version of SAS/STAT do you have? You can verify that with:

proc product_status;run;

Spoiler

@mcraft wrote:

Hello,

I'm trying to fit a model using PROC MCMC and output credible intervals for each parameter. I add the option ods output PostSumInt = intervals, which should output by default. Below is the code I've specified for fitting the model.

ods output Corr = mycorr
Cov = mycov
DIC = mydic
PostSummaries = myparams
Geweke = mygeweke
PostSumInt = intervals; * output ods tables of interest to datasets;
PROC MCMC
data = data SEED=9879 NMC=100000 NBI=500 MAXTUNE=50 THIN=25 STATISTICS = summary DIAGNOSTICS = GEWEKE STATS = CORR STATS = COV DIC;

ARRAY location_scale_mean[2] (0,0);
ARRAY location_scale_cov[2,2] (1,0,0,1);
ARRAY location_scale_rf[2] sub_1 sub_2;

PARMS b0=0.02 b1=0.22 b2=-0.07 b3=0.21 b4=0.13
tau0=0.17 tau1=0.11 tau2=0.13 tau3=0.21 tau4=-0.15
lp0=0.25 lp1=0.12 lp2=-1.08
cov=-0.09 ln_varScale=0.33;

BY rep_n;
WHERE rep_n < 4;

PRIOR b: ~NORMAL (0, VAR=1000);
PRIOR tau: ~NORMAL (0, VAR=1000);
PRIOR lp: ~NORMAL (0, VAR=1000);
PRIOR cov: ~NORMAL (0, VAR=1000);
PRIOR ln_varScale~NORMAL(0,VAR=1000);

varBS = EXP(lp0 + W1_c*lp1 + W2_b*lp2);
varScale = EXP(ln_varScale);

L11=SQRT(varBS);
L21=cov/L11;
L22=SQRT(varScale-L21**2);

mu = b0 + b1*X1_c + b2*X2_b + b3*W1_c + b4*W2_b + L11*sub_1;
varWS = EXP(tau0 + X1_c*tau1 + X2_b*tau2 + W1_c*tau3 + W2_b*tau4 + L21*sub_1 + L22*sub_2);

RANDOM location_scale_rf ~ mvn(location_scale_mean,location_scale_cov) SUBJECT=subj;
MODEL y ~ NORMAL(mu,var = varWS);

RUN;
ods output close;

After running this model, I get an error message that says:

WARNING: Output 'PostSumInt' was not created. Make sure that the output object name, label, or path is spelled correctly. Also,

verify that the appropriate procedure options are used to produce the requested output object. For example, verify that

the NOPRINT option is not used.

Any ideas on why this output was not created? I'm pretty sure it's spelled correctly and I can't find any required options for producing this output.

Thanks!

Madeline

wrote: Hello, I'm trying to fit a model using PROC MCMC and output credible intervals for each parameter. I add the option ods output PostSumInt = intervals, which should output by default. Below is the code I've specified for fitting the model. ods output Corr = mycorrCov = mycovDIC = mydicPostSummaries = myparamsGeweke = mygewekePostSumInt = intervals; * output ods tables of interest to datasets;PROC MCMCdata = data SEED=9879 NMC=100000 NBI=500 MAXTUNE=50 THIN=25 STATISTICS = summary DIAGNOSTICS = GEWEKE STATS = CORR STATS = COV DIC; ARRAY location_scale_mean[2] (0,0);ARRAY location_scale_cov[2,2] (1,0,0,1);ARRAY location_scale_rf[2] sub_1 sub_2; PARMS b0=0.02 b1=0.22 b2=-0.07 b3=0.21 b4=0.13tau0=0.17 tau1=0.11 tau2=0.13 tau3=0.21 tau4=-0.15lp0=0.25 lp1=0.12 lp2=-1.08cov=-0.09 ln_varScale=0.33; BY rep_n;WHERE rep_n < 4; PRIOR b: ~NORMAL (0, VAR=1000);PRIOR tau: ~NORMAL (0, VAR=1000);PRIOR lp: ~NORMAL (0, VAR=1000);PRIOR cov: ~NORMAL (0, VAR=1000);PRIOR ln_varScale~NORMAL(0,VAR=1000); varBS = EXP(lp0 + W1_c*lp1 + W2_b*lp2);varScale = EXP(ln_varScale); L11=SQRT(varBS);L21=cov/L11;L22=SQRT(varScale-L21**2); mu = b0 + b1*X1_c + b2*X2_b + b3*W1_c + b4*W2_b + L11*sub_1;varWS = EXP(tau0 + X1_c*tau1 + X2_b*tau2 + W1_c*tau3 + W2_b*tau4 + L21*sub_1 + L22*sub_2); RANDOM location_scale_rf ~ mvn(location_scale_mean,location_scale_cov) SUBJECT=subj;MODEL y ~ NORMAL(mu,var = varWS); RUN;ods output close; After running this model, I get an error message that says: WARNING: Output 'PostSumInt' was not created. Make sure that the output object name, label, or path is spelled correctly. Also, verify that the appropriate procedure options are used to produce the requested output object. For example, verify that the NOPRINT option is not used. Any ideas on why this output was not created? I'm pretty sure it's spelled correctly and I can't find any required options for producing this output. Thanks! Madeline

mcraft · Posted 12-01-2021 02:13 PM

1.

1 libname s10_100 "C:/Users/mcraft/Desktop/PROC MCMC Simulations/data10_100/output";
NOTE: Libref S10_100 was successfully assigned as follows:
Engine: V9
Physical Name: C:\Users\mcraft\Desktop\PROC MCMC Simulations\data10_100\output

2 ods graphics off; * suppress ods graphics for a more efficient simulation;
3 ods exclude all; * suppress ods output for a more efficient simulation;
4 ods output Corr = s10_100.mycorr
NOTE: Writing HTML Body file: sashtml.htm
5 Cov = s10_100.mycov
6 DIC = s10_100.mydic
7 PostSummaries = s10_100.myparams
8 Geweke = s10_100.mygeweke
9 PostSumInt = s10_100.intervals; * output ods tables of interest to datasets;
10 PROC MCMC
11 data = s10_100.data SEED=9879 NMC=100000 NBI=500 MAXTUNE=50 THIN=25 STATISTICS = summary
11 ! DIAGNOSTICS = GEWEKE STATS = CORR STATS = COV DIC;
12 *data = s10_100.data SEED=9879 NMC=100000 NBI=500 MAXTUNE=50 THIN=25 NTHREADS=4 DIAGNOSTICS =
12 ! AUTOCORR PROPCOV = method(ITPRINT) STATS = CORR STATS = COV DIC;
13
14 ARRAY location_scale_mean[2] (0,0);
15 ARRAY location_scale_cov[2,2] (1,0,0,1);
16 ARRAY location_scale_rf[2] sub_1 sub_2;
17
18 PARMS b0=0.02 b1=0.22 b2=-0.07 b3=0.21 b4=0.13
19 tau0=0.17 tau1=0.11 tau2=0.13 tau3=0.21 tau4=-0.15
20 lp0=0.25 lp1=0.12 lp2=-1.08
21 cov=-0.09 ln_varScale=0.33;
22
23 BY rep_n;
24 WHERE rep_n = 1;
25
26 PRIOR b: ~NORMAL (0, VAR=1000);
27 PRIOR tau: ~NORMAL (0, VAR=1000);
28 PRIOR lp: ~NORMAL (0, VAR=1000);
29 PRIOR cov: ~NORMAL (0, VAR=1000);
30 PRIOR ln_varScale~NORMAL(0,VAR=1000);
31
32 varBS = EXP(lp0 + W1_c*lp1 + W2_b*lp2);
33 varScale = EXP(ln_varScale);
34
35 L11=SQRT(varBS);
36 L21=cov/L11;
37 L22=SQRT(varScale-L21**2);
38
39 mu = b0 + b1*X1_c + b2*X2_b + b3*W1_c + b4*W2_b + L11*sub_1;
40 varWS = EXP(tau0 + X1_c*tau1 + X2_b*tau2 + W1_c*tau3 + W2_b*tau4 + L21*sub_1 + L22*sub_2);
41
42 RANDOM location_scale_rf ~ mvn(location_scale_mean,location_scale_cov) SUBJECT=subj;
43 MODEL y ~ NORMAL(mu,var = varWS);
44
45 RUN;

NOTE: Tuning the proposal distribution.
NOTE: Generating the burn-in samples.
NOTE: Beginning sample generation.
NOTE: The above message was for the following BY group:
rep_n=1
NOTE: The data set S10_100.MYGEWEKE has 15 observations and 4 variables.
NOTE: The data set S10_100.MYPARAMS has 15 observations and 8 variables.
NOTE: The data set S10_100.MYDIC has 4 observations and 3 variables.
NOTE: The data set S10_100.MYCOV has 15 observations and 17 variables.
NOTE: The data set S10_100.MYCORR has 15 observations and 17 variables.
WARNING: Output 'PostSumInt' was not created. Make sure that the output object name, label, or path
is spelled correctly. Also, verify that the appropriate procedure options are used to
produce the requested output object. For example, verify that the NOPRINT option is not
used.
NOTE: PROCEDURE MCMC used (Total process time):
real time 1:28.96
cpu time 1:28.48

46 ods output close;
47 ods exclude none;

2.

48 proc product_status;run;

For Base SAS Software ...
Custom version information: 9.4
Image version information: 9.04.01M0P061913
For SAS/STAT ...
Custom version information: 12.3 (maintenance)
For SAS/GRAPH ...
Custom version information: 9.4
For SAS/ETS ...
Custom version information: 12.3 (maintenance)
For SAS/FSP ...
Custom version information: 9.4
For SAS/OR ...
Custom version information: 12.3 (maintenance)
For SAS/AF ...
Custom version information: 9.4
For SAS/IML ...
Custom version information: 12.3 (maintenance)
For SAS/QC ...
Custom version information: 12.3 (maintenance)
For SAS/SHARE ...
Custom version information: 9.4
For SAS/ASSIST ...
Custom version information: 9.4
For SAS/CONNECT ...
Custom version information: 9.4
For SAS/TOOLKIT ...
Custom version information: 9.4
For SAS/EIS ...
Custom version information: 9.4
For SAS/GIS ...
Custom version information: 9.4
For SAS/ACCESS Interface to Netezza ...
Custom version information: 9.4
For SAS Enterprise Miner ...
Custom version information: 12.3
For SAS/ACCESS to Vertica ...
Custom version information: 9.4
For SAS/ACCESS to Postgres ...
Custom version information: 9.4
For SAS/IntrNet ...
Custom version information: 9.4
For SAS Integration Technologies ...
Custom version information: 9.4
For SAS Text Miner ...
Custom version information: 12.3
For SAS/ACCESS to SAP HANA ...
Custom version information: 9.4
Image version information: 9.04.01M0P071013
For SAS/Genetics ...
Custom version information: 9.4
For SAS/ACCESS Interface to DB2 ...
Custom version information: 9.4
For SAS/ACCESS Interface to Oracle ...
Custom version information: 9.4
For SAS/ACCESS Interface to Sybase ...
Custom version information: 9.4
For SAS/ACCESS Interface to PC Files ...
Custom version information: 9.4
For SAS/ACCESS Interface to ODBC ...
Custom version information: 9.4
For SAS/ACCESS Interface to OLE DB ...
Custom version information: 9.4
For SAS/ACCESS Interface to Teradata ...
Custom version information: 9.4
For SAS/ACCESS Interface to MySQL ...
Custom version information: 9.4

Reeza · Posted 12-01-2021 03:03 PM

You're using an old version of SAS/STAT, the current version is 15, you're on 12.3.
The 12.3 documentation doesn't have a PostSumInt table listed, would any of the other options get you there?
https://support.sas.com/documentation/cdl/en/statug/66103/HTML/default/viewer.htm#statug_mcmc_detail...

It looks like this isn't a feature in your current version 😞

mcraft · Posted 12-01-2021 04:56 PM

Thank you so much! I will see if I can get it updated.

Reeza · Posted 12-01-2021 05:01 PM

It looks like you're running an 'old' version, 9.4M0 which was released in 2013
https://blogs.sas.com/content/iml/2013/08/02/how-old-is-your-version-of-sas-release-dates-for-sas-so...

mcraft · Posted 12-01-2021 06:25 PM

I have a follow-up issue. I'm now using the ODS Table Name PostIntervals since this option outputs credible intervals for 12.3. Below is my model code, but I'm still getting an error message (see below).

ods graphics off; * suppress ods graphics for a more efficient simulation;
ods exclude all; * suppress ods output for a more efficient simulation;
ods output Corr = s10_100.mycorr
Cov = s10_100.mycov
DIC = s10_100.mydic
PostSummaries = s10_100.myparams
Geweke = s10_100.mygeweke
PostIntervals = s10_100.intervals; * output ods tables of interest to datasets;
PROC MCMC
data = s10_100.data SEED=9879 NMC=100000 NBI=500 MAXTUNE=50 THIN=25 STATISTICS = summary DIAGNOSTICS = GEWEKE STATS = CORR STATS = COV DIC;
*data = s10_100.data SEED=9879 NMC=100000 NBI=500 MAXTUNE=50 THIN=25 NTHREADS=4 DIAGNOSTICS = AUTOCORR PROPCOV = method(ITPRINT) STATS = CORR STATS = COV DIC;

ARRAY location_scale_mean[2] (0,0);
ARRAY location_scale_cov[2,2] (1,0,0,1);
ARRAY location_scale_rf[2] sub_1 sub_2;

PARMS b0=0.02 b1=0.22 b2=-0.07 b3=0.21 b4=0.13
tau0=0.17 tau1=0.11 tau2=0.13 tau3=0.21 tau4=-0.15
lp0=0.25 lp1=0.12 lp2=-1.08
cov=-0.09 ln_varScale=0.33;

BY rep_n;
WHERE rep_n < 4;

PRIOR b: ~NORMAL (0, VAR=1000);
PRIOR tau: ~NORMAL (0, VAR=1000);
PRIOR lp: ~NORMAL (0, VAR=1000);
PRIOR cov: ~NORMAL (0, VAR=1000);
PRIOR ln_varScale~NORMAL(0,VAR=1000);

varBS = EXP(lp0 + W1_c*lp1 + W2_b*lp2);
varScale = EXP(ln_varScale);

L11=SQRT(varBS);
L21=cov/L11;
L22=SQRT(varScale-L21**2);

mu = b0 + b1*X1_c + b2*X2_b + b3*W1_c + b4*W2_b + L11*sub_1;
varWS = EXP(tau0 + X1_c*tau1 + X2_b*tau2 + W1_c*tau3 + W2_b*tau4 + L21*sub_1 + L22*sub_2);

RANDOM location_scale_rf ~ mvn(location_scale_mean,location_scale_cov) SUBJECT=subj;
MODEL y ~ NORMAL(mu,var = varWS);

RUN;
ods output close;
ods exclude none;

NOTE: Tuning the proposal distribution.
NOTE: Generating the burn-in samples.
NOTE: Beginning sample generation.
NOTE: The above message was for the following BY group:
rep_n=1
NOTE: Tuning the proposal distribution.
NOTE: Generating the burn-in samples.
NOTE: Beginning sample generation.
NOTE: The above message was for the following BY group:
rep_n=2
NOTE: Tuning the proposal distribution.
NOTE: Generating the burn-in samples.
NOTE: Beginning sample generation.
NOTE: The above message was for the following BY group:
rep_n=3
NOTE: The data set S10_100.MYGEWEKE has 45 observations and 4 variables.
NOTE: The data set S10_100.MYPARAMS has 45 observations and 8 variables.
NOTE: The data set S10_100.MYDIC has 12 observations and 3 variables.
NOTE: The data set S10_100.MYCOV has 45 observations and 17 variables.
NOTE: The data set S10_100.MYCORR has 45 observations and 17 variables.
WARNING: Output 'PostIntervals' was not created. Make sure that the output object name, label, or
path is spelled correctly. Also, verify that the appropriate procedure options are used to
produce the requested output object. For example, verify that the NOPRINT option is not
used.
NOTE: PROCEDURE MCMC used (Total process time):
real time 4:24.27
cpu time 4:23.96

Reeza · Posted 12-01-2021 06:47 PM

Try moving your ODS OUTPUT statement to within the PROC but I don't see anything that should cause that issue.
Given the version you'll have to contact SAS tech support for questions beyond this as we can't test anything.

mcraft · Posted 12-01-2021 07:17 PM

If I were to use SAS on Demand, the PostSumInt option should work because it'll be the most up-to-date version of SAS, right?

I ask because I'm getting the same error message using SAS on Demand, which makes me wonder whether the issue is broader than just the SAS version I'm working with.

Thank you for all of your help!

Reeza · Posted 12-01-2021 07:39 PM

PostSumInt works fine on On Demand for me - example from documentation was used.

/*-----------------------------------------------------------------
            S A S   S A M P L E   L I B R A R Y                   
                                                                  
      NAME: MCMCEX4                                               
     TITLE: Documentation Example 4 for PROC MCMC                 
            Logistic Regression Model with Jeffreys' Prior        
   PRODUCT: STAT                                                  
    SYSTEM: ALL                                                   
      KEYS:                                                       
     PROCS: MCMC                                                  
      DATA:                                                       
                                                                  
   SUPPORT: Fang Chen                                             
       REF: PROC MCMC, EXAMPLE 4                                  
      MISC:                                                       
-----------------------------------------------------------------*/

title 'Logistic Regression Model with Jeffreys Prior';
data vaso;
   input vol rate resp @@;
   lvol = log(vol);
   lrate = log(rate);
   ind = _n_;
   cnst = 1;
   datalines;
3.7  0.825  1  3.5  1.09  1  1.25  2.5   1  0.75  1.5  1
0.8  3.2    1  0.7  3.5   1  0.6   0.75  0  1.1   1.7  0
0.9  0.75   0  0.9  0.45  0  0.8   0.57  0  0.55  2.75 0
0.6  3.0    0  1.4  2.33  1  0.75  3.75  1  2.3  1.64  1
3.2  1.6    1  0.85 1.415 1  1.7   1.06  0  1.8  1.8   1
0.4  2.0    0  0.95 1.36  0  1.35  1.35  0  1.5  1.36  0
1.6  1.78   1  0.6  1.5   0  1.8   1.5   1  0.95 1.9   0
1.9  0.95   1  1.6  0.4   0  2.7   0.75  1  2.35 0.03  0
1.1  1.83   0  1.1  2.2   1  1.2   2.0   1  0.8  3.33  1
0.95 1.9    0  0.75 1.9   0  1.3   1.625 1
;

%let n = 39;
proc mcmc data=vaso nmc=10000 outpost=mcmcout seed=17;
   ods output PostSumInt = want;

   array beta[3] beta0 beta1 beta2;
   array m[&n, &n];
   array x[1] / nosymbols;
   array xt[3, &n];
   array xtm[3, &n];
   array xmx[3, 3];
   array p[&n];

   parms beta0 1 beta1 1 beta2 1;

   begincnst;
      if (ind eq 1) then do;
         rc = read_array("vaso", x, "cnst", "lvol", "lrate");
         call transpose(x, xt);
         call zeromatrix(m);
      end;
   endcnst;

   beginnodata;
   call mult(x, beta, p);              /* p = x * beta */
   do i = 1 to &n;
      p[i] = 1 / (1 + exp(-p[i]));     /* p[i] = 1/(1+exp(-x*beta)) */
      m[i,i] = p[i] * (1-p[i]);
   end;
   call mult (xt, m, xtm);             /* xtm = xt * m        */
   call mult (xtm, x, xmx);            /* xmx = xtm * x       */
   call det (xmx, lp);                 /* lp = det(xmx)       */
   lp = 0.5 * log(lp);                 /* lp = -0.5 * log(lp) */
   prior beta: ~ general(lp);
   endnodata;

   model resp ~ bern(p[ind]);
run;

Log:

 NOTE: The input data set variable ind appears in the distributional specification for beta0. This might lead to incorrect results.
 NOTE: Tuning the proposal distribution.
 NOTE: Generating the burn-in samples.
 NOTE: Beginning sample generation.
 NOTE: Generating diagnostic plots.
 NOTE: The data set WORK.WANT has 3 observations and 6 variables.
 NOTE: The data set WORK.MCMCOUT has 10000 observations and 7 variables.
 NOTE: PROCEDURE MCMC used (Total process time):
       real time           1.13 seconds
       user cpu time       0.71 seconds
       system cpu time     0.08 seconds
       memory              24393.70k
       OS Memory           51800.00k
       Timestamp           12/02/2021 12:36:09 AM
       Step Count                        114  Switch Count  41
       Page Faults                       0
       Page Reclaims                     11234
       Page Swaps                        0
       Voluntary Context Switches        761
       Involuntary Context Switches      1
       Block Input Operations            0
       Block Output Operations           17288

@mcraft wrote:

If I were to use SAS on Demand, the PostSumInt option should work because it'll be the most up-to-date version of SAS, right?

I ask because I'm getting the same error message using SAS on Demand, which makes me wonder whether the issue is broader than just the SAS version I'm working with.

Thank you for all of your help!

If you're having further issues it could be related to your data or something else. If you have a licensed version of SAS I'd call Tech Support at this point.

Ready to join fellow brilliant minds for the SAS Hackathon?

Classroom Training Available!