Multiple processors, CFA, imputed dat... PreviousNext
Mplus Discussion > Confirmatory Factor Analysis >
Message/Author
 Baljinder Sahdra posted on Tuesday, July 09, 2013 - 4:52 pm
I'm running a large CFA with ML estimator, 300+ indicators of dozens of factors and 7800+ cases, using 25 imputed files. (The data were collected using missing-at-random design.)

I am using Mplus Version 7 64-bit. My PC has Intel i7 with 12 cores, 64 GB RAM, and Windows 7 Enterprise.

To speed up processing, I tried using processors=12 (or 8 or 6), but no matter what value I use, Mplus won't use more than about 8% CPU and about 4 cores. If I don't use the processors command at all, it still uses about 8% CPU and about 4 cores. I also tried specifying STARTS, but no change.

I've tried using several different types to analyses to see if maybe the processors command is not available with imputed files analysis, but it’s the same story even while running a plain old multi-factor CFA on a single data file.

All 12 cores of my machine seem to work fine (as confirmed by an independent stress test on them), but Mplus is not making the most of them.

Any ideas how I might make Mplus use more of my computer's computational power than it seems to be using?

Thank you in advance for your guidance.
 Linda K. Muthen posted on Wednesday, July 10, 2013 - 12:11 pm
The PROCESSORS option is not available for all analyses. See the PROCESSORS option in the user's guide for a description of when it is available.
 Baljinder Sahdra posted on Wednesday, July 10, 2013 - 2:01 pm
Thank you for your reply. I am using maximum likelihood estimation for a CFA using all continuous variables. Based on the information in the user's guide (pasted below), I assumed that the PROCESSORS option should be available for my model. But Mplus is not using all 12 cores for this kind of analysis with either imputed dataset or a single data file. Is there anything else I can do?


From the user's guide (p. 644):
MULTIPLE PROCESSORS
The use of multiple processors without threads is available for TYPE=MIXTURE; Bayesian analysis with more than one chain unless STVALUES=ML; models that require numerical integration; models with all continuous variables, missing data, and maximum likelihood
estimation; and TYPE=TWOLEVEL with categorical outcomes and ESTIMATOR= WLSMV.
 Linda K. Muthen posted on Wednesday, July 10, 2013 - 2:35 pm
What version of Mplus are you using?
 Baljinder Sahdra posted on Wednesday, July 10, 2013 - 2:50 pm
I am using Mplus Version 7 (64-bit).
 Linda K. Muthen posted on Wednesday, July 10, 2013 - 5:21 pm
Please send your input, data, and license number to support@statmodel.com.
 Jonathon Little posted on Friday, January 10, 2014 - 2:57 am
Regarding the scalability of Mplus. What is the maximum number of processor cores that can be used for the type of the analyses listed above. Is there a limit?

Many thanks.
 Bengt O. Muthen posted on Friday, January 10, 2014 - 4:02 pm
No limit.
 j guo posted on Wednesday, May 28, 2014 - 4:17 pm
Hi, Dr. Muthen

I tried to impute multiple datasets for a set of variables with missing values using sampling weights, cluster and grouping (see below)

variables: x1-x11;
cluster = CONTSCHL;
WEIGHT = W_FSTU;
GROUPING = GROUP (1 2 3 4 5);
define: CONTSCHL=(COUNTRY*10000)+SCHOOLID;

DATA IMPUTATION:
impute = x1-x11;
NDATASETS =5;
SAVE = imp*.dat;
Analysis: etimator=MLR; TYPE=COMPLEX:

I got the imputed dataset and found MPLUS create a new weights variable for me.
What is the difference between new sampling weights and original weights (i.e., W_FSTU) ?
 Tihomir Asparouhov posted on Wednesday, May 28, 2014 - 9:07 pm
The weights are scaled to sum up to the sample size. This is not related to the imputation.
Back to top
Add Your Message Here
Post:
Username: Posting Information:
This is a private posting area. Only registered users and moderators may post messages here.
Password:
Options: Enable HTML code in message
Automatically activate URLs in message
Action: