Hello!
I can't run bootstrapping analysis with 5000 resamples.... it takes a long time and when it's getting to 3200 resamples, it stops. I tried it several times and it doesn't work with me. I talked with some collegues and they seem to have had similar problems before. I was wondering if someone knows what might be the problem?
Eliana
(Portuguese PhD)
running bootstrapping with 5000 resamples
-
- PLS User
- Posts: 13
- Joined: Mon Oct 13, 2008 5:30 pm
- Real name and title:
-
- PLS User
- Posts: 13
- Joined: Mon Oct 13, 2008 5:30 pm
- Real name and title:
running 5000 bootstraps
Well, previous studies usually run 5000 bootstraps, although they're not PLS studies. Moreover, there is no specific number of resamples, but as a general rule of thumb studies recommend running at least 1000. But, of course, to make more valid and reliable, the more bootstrap resamples you can get, the better it will be.
And, PLS does run 5000 bootstraps, but with simple models. Mine has a lot of latent variables and paths and i think that is why it is running out of "memory".
I'm just wondering why you said it didn't make sense to use 5000 bootstraps.
Eliana
And, PLS does run 5000 bootstraps, but with simple models. Mine has a lot of latent variables and paths and i think that is why it is running out of "memory".
I'm just wondering why you said it didn't make sense to use 5000 bootstraps.
Eliana
- Diogenes
- PLS Super-Expert
- Posts: 899
- Joined: Sat Oct 15, 2005 5:13 pm
- Real name and title:
- Location: São Paulo - BRAZIL
- Contact:
Hi Eliana,
in fact I haven't new information, and I don't know if this kind of problem is a hardware, software or a model problem.
More replications are better, and one idea is:
- Run the bootstrap procedure with 200 replication and save the results
- Run with 500 replication and save the results
- Run with 1000 replication and save the results
- Run with 1500 replication and save the results
- Compare all results. Probably all coefficients that were significant in the 1st run will be significant in all others. It’s a way to be more confident about the stability of the results.
Best regards,
Bido
in fact I haven't new information, and I don't know if this kind of problem is a hardware, software or a model problem.
More replications are better, and one idea is:
- Run the bootstrap procedure with 200 replication and save the results
- Run with 500 replication and save the results
- Run with 1000 replication and save the results
- Run with 1500 replication and save the results
- Compare all results. Probably all coefficients that were significant in the 1st run will be significant in all others. It’s a way to be more confident about the stability of the results.
Best regards,
Bido
-
- PLS User
- Posts: 13
- Joined: Mon Oct 13, 2008 5:30 pm
- Real name and title:
A possible solution to run 5000 bootstraps
Following a suggestion I found here on the forum, I was able to run 5000 bootstraps without any problem. It was really fast and easy.
The problem was related with the virtual memory required for java to work and run the models.
As Alexander Will suggested, "for the calculation of very large models it could be necessary to tell the Java Virtual Machine that it can use more memory that it would per default".
However, when I tried to use his command, it did not work, so I had to find a new one.
To increase the virtual memory, you can do it manually by starting SmartPLS from the command shell.
To find the command shell, press windows symbol on your board + r
Then, write 'cmd' on the box and press enter
Then, write cd\ + enter
Then, write cd program files + enter
Then, write cd smartpls2 + enter
And finally, write the following: java -Xms512M -Xmx1024M -jar startup.jar
Press enter and smartpls will open. Then try to run your model :))
Hope this helps.
Eliana
The problem was related with the virtual memory required for java to work and run the models.
As Alexander Will suggested, "for the calculation of very large models it could be necessary to tell the Java Virtual Machine that it can use more memory that it would per default".
However, when I tried to use his command, it did not work, so I had to find a new one.
To increase the virtual memory, you can do it manually by starting SmartPLS from the command shell.
To find the command shell, press windows symbol on your board + r
Then, write 'cmd' on the box and press enter
Then, write cd\ + enter
Then, write cd program files + enter
Then, write cd smartpls2 + enter
And finally, write the following: java -Xms512M -Xmx1024M -jar startup.jar
Press enter and smartpls will open. Then try to run your model :))
Hope this helps.
Eliana
Re: running bootstrapping with 5000 resamples
Does this method also work for macbook in smartpls 3?