Problem with estimating propose matrix
-
- Posts: 11
- Joined: March 28 2012
- Affiliation: Institute for Theoretical Physics, University of Zurich
Problem with estimating propose matrix
Hi,
I have written a module about galaxy cluster counts and clustering and trying to do constraint on cosmological parameters. My code gives me chi_square for given cosmology. For the first run, I am trying to estimate the propose matrix by setting "estimate_propose_matrix = T" and "propose_matrix as blank". I am facing some problems in this as follows:
1. When I am initializing the chains, it is generating file_root files in the specified location, but not writing anything to file_root.txt files.
2. The chains are continue to run even after the specified number of maximum steps has been reached. For example, I set up the maximum steps to be around 10000 whereas my chains are still running after more than 15000 steps. I got to know about the number of steps from the file_root.log files (camb called xxxx times).
3. It seems to me that the chains are running forever and is not doing anything to txt files.
Can somebody please suggest me how can I get my .covmat file in this scenario so that I can run the chains with the specified propose matrix? Are there some flags that need to be taken care of?
I am compiling with mpi.
Any help is very much appreciated. Thanks in advance.
Irshad
I have written a module about galaxy cluster counts and clustering and trying to do constraint on cosmological parameters. My code gives me chi_square for given cosmology. For the first run, I am trying to estimate the propose matrix by setting "estimate_propose_matrix = T" and "propose_matrix as blank". I am facing some problems in this as follows:
1. When I am initializing the chains, it is generating file_root files in the specified location, but not writing anything to file_root.txt files.
2. The chains are continue to run even after the specified number of maximum steps has been reached. For example, I set up the maximum steps to be around 10000 whereas my chains are still running after more than 15000 steps. I got to know about the number of steps from the file_root.log files (camb called xxxx times).
3. It seems to me that the chains are running forever and is not doing anything to txt files.
Can somebody please suggest me how can I get my .covmat file in this scenario so that I can run the chains with the specified propose matrix? Are there some flags that need to be taken care of?
I am compiling with mpi.
Any help is very much appreciated. Thanks in advance.
Irshad
-
- Posts: 1944
- Joined: September 23 2004
- Affiliation: University of Sussex
- Contact:
Re: Problem with estimating propose matrix
estimate_propose_matrix is not very reliable.
If neccessary you can start without a covmat (using MPI learning options), or usually better to start with one of the supplied covmat files - it doesn't matter if you are varying additional parameters as long as you set the ranges for the additional parameters approximately in the .ini file (and again use MPI learning option).
If neccessary you can start without a covmat (using MPI learning options), or usually better to start with one of the supplied covmat files - it doesn't matter if you are varying additional parameters as long as you set the ranges for the additional parameters approximately in the .ini file (and again use MPI learning option).
-
- Posts: 1944
- Joined: September 23 2004
- Affiliation: University of Sussex
- Contact:
Re: Problem with estimating propose matrix
As of the Oct 2012 version, best-fit finding (action=2) and estimate_propose_matrix should work more reliably, esp. with parameters that have hard prior cuts.
-
- Posts: 11
- Joined: March 28 2012
- Affiliation: Institute for Theoretical Physics, University of Zurich
Problem with estimating propose matrix
Thank you so much Antony for the update.