I have inadvertently found out that the SEM modelling I am doing is far slower using the SEM builder rather than the command prompt. This can be replicated using:
1) Open a stata datafile
2) Open a stata SEM builder file
3) Click ‘evaluate’ to run the SEM code. <-- time the time it takes to calculate
4) Copy the code from the command prompt in step 3 to the clipboard
5) Type clear all
6) Reopen the stata datafile
7) Paste the code to the stata dialog box <-- time the time it takes to calculate
For me for a very simple code, the timings in step 3 was 25 seconds, and step 7 was 2.9 seconds. This 25 seconds is limited not by processing power (I get similar results using SE/MP2/MP4).
This is with 300 variables, several latent variables, and mlmv for missing data. I have been steadily upgrading stata to MP to try and get a faster response, but it seems the limiting factor seems to be in the SEM builder (?bug) - I should be using the command prompt for fastest responses. I assume this is because each iteration etc. is being fed back to the SEM builder graphical interface, so slows things down. But there might be other reasons ?RAM due to graphical interface. I have gone back and tried Stata 16 MP2, and the same thing happens.
Any thoughts? Anyone else noticed this?
1) Open a stata datafile
2) Open a stata SEM builder file
3) Click ‘evaluate’ to run the SEM code. <-- time the time it takes to calculate
4) Copy the code from the command prompt in step 3 to the clipboard
5) Type clear all
6) Reopen the stata datafile
7) Paste the code to the stata dialog box <-- time the time it takes to calculate
For me for a very simple code, the timings in step 3 was 25 seconds, and step 7 was 2.9 seconds. This 25 seconds is limited not by processing power (I get similar results using SE/MP2/MP4).
This is with 300 variables, several latent variables, and mlmv for missing data. I have been steadily upgrading stata to MP to try and get a faster response, but it seems the limiting factor seems to be in the SEM builder (?bug) - I should be using the command prompt for fastest responses. I assume this is because each iteration etc. is being fed back to the SEM builder graphical interface, so slows things down. But there might be other reasons ?RAM due to graphical interface. I have gone back and tried Stata 16 MP2, and the same thing happens.
Any thoughts? Anyone else noticed this?
Comment