Independts of chains?
-
- Posts: 14
- Joined: November 06 2013
- Affiliation: CMU
Independts of chains?
Hello All,
Is it technically/scientifically right to run cosmomc several time with one chain and combine the chains while using getdit to get the marginalized likelihood?
I can run hundreds of single processor job but I can't run parallel job on many processor due to some technical difficulty.
Sorry if its a repetition, I couldn't find any discussion on the topic in cosmocoffee arxives.
Thanks,
Shadab
Is it technically/scientifically right to run cosmomc several time with one chain and combine the chains while using getdit to get the marginalized likelihood?
I can run hundreds of single processor job but I can't run parallel job on many processor due to some technical difficulty.
Sorry if its a repetition, I couldn't find any discussion on the topic in cosmocoffee arxives.
Thanks,
Shadab
-
- Posts: 97
- Joined: March 19 2010
- Affiliation: The University of Texas at Dallas
- Contact:
Independts of chains?
You can actually run numbered single chain jobs using for example
This will produce numbered chains just like if you had run in MPI. Of course while the jobs are running they wont crosstalk. You will simply have a purely MCMC sampling of the parameter space.
You can then run getdist on these chains in the usual way, set the number of chains you ran by setting the chain_num parameter in your getdist ini file.
Code: Select all
./cosmomc test.ini 1
./cosmomc test.ini 2
You can then run getdist on these chains in the usual way, set the number of chains you ran by setting the chain_num parameter in your getdist ini file.
-
- Posts: 14
- Joined: November 06 2013
- Affiliation: CMU
Independts of chains?
Thanks Jason,
That helps a lot.
Could you clarify another thing related to proposal update.
Since these chains will be completely independent including the proposal update. Running chains independently is going to bias the results or make things very inefficient for convergence?
Thanks,
Shadab
That helps a lot.
Could you clarify another thing related to proposal update.
Since these chains will be completely independent including the proposal update. Running chains independently is going to bias the results or make things very inefficient for convergence?
Thanks,
Shadab
-
- Posts: 97
- Joined: March 19 2010
- Affiliation: The University of Texas at Dallas
- Contact:
Independts of chains?
Running the chains independently should not bias your results, rather, as you said, it will just be very inefficient in achieving convergence.
-
- Posts: 14
- Joined: November 06 2013
- Affiliation: CMU
Independts of chains?
Thanks Jason
I appreciate your help :)
Shadab
I appreciate your help :)
Shadab