CosmoMC: finding max like point

Use of Cobaya. camb, CLASS, cosmomc, compilers, etc.
Post Reply
Davide Maino
Posts: 11
Joined: September 22 2005
Affiliation: Physics Dept., University of Milano

CosmoMC: finding max like point

Post by Davide Maino » September 06 2006

Hi Anthony,
I'm trying to run CosmoMC with all_exact_l and set estimate_propose_matrix = T. After a while the code exits with and error

ERROR: Trial parameters excluded by prior or error in likelihood
Try starting further away from problem regions?

I have two questions:
1. how can I solve the problem?
2. Wouldn't it be nice to have indication from the code which is the parameters that causes the error?

Thanks a lot,

- Davide.

Antony Lewis
Posts: 1941
Joined: September 23 2004
Affiliation: University of Sussex
Contact:

Re: CosmoMC: finding max like point

Post by Antony Lewis » September 07 2006

I never use this option myself.. as people have seen it is not very reliable, especially with parameters with priors. If anyone wants to improve it I'd be happy to incorporate improvements...

btw, someone found a bug in August 2006 CosmoMC version with tensor modes. I updated the download today.

Davide Maino
Posts: 11
Joined: September 22 2005
Affiliation: Physics Dept., University of Milano

CosmoMC: finding max like point

Post by Davide Maino » September 07 2006

Hi Anthony,
thanks for the prompt answer. I'll download the new version when ready.
So you suggest me not to use the option and let things go by themselves.

I'm quite new to these issues but given the nature of the simulations
would it be faster to first get an idea of the propose_matrix so that the
MCMC part would better concentrate on the right parameter space region?

Ciao,


- Davide.

Antony Lewis
Posts: 1941
Joined: September 23 2004
Affiliation: University of Sussex
Contact:

Re: CosmoMC: finding max like point

Post by Antony Lewis » September 07 2006

If you can get some propose_matrix estimate that will help. But if you don't have one the learn_propose option usually works quite well with sensible input parameters (and is a lot more robust that the estimate_propose_matrix option).

Davide Maino
Posts: 11
Joined: September 22 2005
Affiliation: Physics Dept., University of Milano

CosmoMC: finding max like point

Post by Davide Maino » September 07 2006

OK. Fine for me.
Ciao,


- Davide.

Jie Ren
Posts: 28
Joined: May 13 2006
Affiliation: Theoretical Physics Division, Chern Institute of Mathematics, Nankai University, Tianjin, China

CosmoMC: finding max like point

Post by Jie Ren » September 23 2006

I do not know how to solve the following problem about propose matrix. Thanks a lot for any helps!

In CosmoMC Readme,
*****************************************
If you don't have a propose_matrix, set estimate_propose_matrix = T to automatically esimate it by numerical differentiation about the best fit point. To use this option you should not have parameter priors cutting off the distribution near the maximum likelihood.
*****************************************

When I set estimate_propose_matrix=T and run, why does this error occur:

[root@localhost cosmomc]# ./cosmomc params.ini
Random seeds: 22217, 28465 rand_inst: 0
Cannot have estimate_propose_matrix and propose_matrix
[root@localhost cosmomc]#

And what is the meaning of the matrix in params_CMB.covmat? If one adds some parameters, how can one deal with the problem of propose matrix?

Jie Ren
Posts: 28
Joined: May 13 2006
Affiliation: Theoretical Physics Division, Chern Institute of Mathematics, Nankai University, Tianjin, China

CosmoMC: finding max like point

Post by Jie Ren » September 24 2006

I have known that this error is due to that the line "propose_matrix = params_CMB.covmat" should be commented when one sets "estimate_propose_matrix = T".

Sarah Bridle
Posts: 144
Joined: September 24 2004
Affiliation: University College London (UCL)
Contact:

CosmoMC: finding max like point

Post by Sarah Bridle » September 29 2006

Is the error solved on commenting out the line
"propose_matrix = params_CMB.covmat"
?

Jie Ren
Posts: 28
Joined: May 13 2006
Affiliation: Theoretical Physics Division, Chern Institute of Mathematics, Nankai University, Tianjin, China

CosmoMC: finding max like point

Post by Jie Ren » September 29 2006

Commenting the line "propose_matrix = params_CMB.covmat" seems equal to modify it as "propose_matrix = ". If one adds parameters and also uses "propose_matrix = params_CMB.covmat", this error will occur. Sometimes, this error is due to not running "make clean" after one modifies some files of the program.

Antony Lewis
Posts: 1941
Joined: September 23 2004
Affiliation: University of Sussex
Contact:

Re: CosmoMC: finding max like point

Post by Antony Lewis » October 05 2012

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.

Post Reply