CosmoCoffee Forum Index CosmoCoffee

 
 FAQFAQ   SearchSearch  MemberlistSmartFeed   MemberlistMemberlist    RegisterRegister 
   ProfileProfile   Log inLog in 
Arxiv New Filter | Bookmarks & clubs | Arxiv ref/author:

Analysis After Increasing the temperature of the chains in commomn.ini
 
Post new topic   Reply to topic    CosmoCoffee Forum Index -> Computers and software
View previous topic :: View next topic  
Author Message
Mayukh Gangopadhyay



Joined: 04 Aug 2017
Posts: 6
Affiliation: Saha Institute of Nuclear Physics

PostPosted: August 23 2017  Reply with quote

Hi,
I am dealing with a very slowly converging simulation. The reason I realised is a long tail and thus the slow mixing. So as advised in the cosmomc readme file, I increased the temperature of the chains from 1 to 2. Now my runs are much faster and one of the runs already converged to the desired R−1 value. Now my question is how to readjust the statistics obtained from this run to get the true statistical values for the parameters in question?
In the readme file it says, GetDist can adjust for the temperature automatically. But I am not sure where or how it does.
Another thing regarding this problem is: I understand increasing the temperature means the T is increased in P(x)1/T. Now is there any upper limit that one should use when you have to increase T. What I mean, is taking T=2 too much of an increase?
It would be very nice to get some references where T is being increased to do the analysis.
Back to top
View user's profile  
Antony Lewis



Joined: 23 Sep 2004
Posts: 1273
Affiliation: University of Sussex

PostPosted: August 23 2017  Reply with quote

You just need to set the "cool" parameter when running getdist, e.g. cool=2 in your case. (adjusting for temperature is easy as long as the temperature is not too high: effectively it importance samples from the high temperature distribution to the low-temperature one).
Back to top
View user's profile [ Hidden ] Visit poster's website
Antony Lewis



Joined: 23 Sep 2004
Posts: 1273
Affiliation: University of Sussex

PostPosted: August 23 2017  Reply with quote

PS. I suggest never editing files in batch2/ etc. The way to change parameters is to make a new ini file in another directory (like the supplied test.ini), which uses DEFAULT(..filename...) to includes settings you want from from files in batch2. You can then inherit and override settings in batch2 without changing the original files.
Back to top
View user's profile [ Hidden ] Visit poster's website
Mayukh Gangopadhyay



Joined: 04 Aug 2017
Posts: 6
Affiliation: Saha Institute of Nuclear Physics

PostPosted: August 23 2017  Reply with quote

Thank you Anthony for your prompt reply.
Back to top
View user's profile  
Mayukh Gangopadhyay



Joined: 04 Aug 2017
Posts: 6
Affiliation: Saha Institute of Nuclear Physics

PostPosted: August 25 2017  Reply with quote

Hi,
I have another problem regarding the same issue. I state it here:

After running with temperature = 2, when we analyse the data with cool = 2 in distparams.ini with "python python/GetDist.py distparams.ini", following is the error that we got:

Best fit sample -log(Like) = 1266.571000

mean(-Ln(like)) = 1272.535056

-Ln(mean like) = 1268.978640

Traceback (most recent call last):

File "python/GetDist.py", line 357, in <module>
main(args)
File "python/GetDist.py", line 194, in main
mc._setDensitiesandMarge1D(writeDataToFile=not no_plots and plot_data_dir, meanlikes=plot_meanlikes)
File "CosmoMC-Nov2016_08/python/getdist/mcsamples.py", line 2077, in _setDensitiesandMarge1D
meanlikes=meanlikes)
File "CosmoMC-Nov2016_08/python/getdist/mcsamples.py", line 1281, in get1DDensityGridData
* (binmax - binmin) * abs(smooth_scale_1D) / fine_width
File "CosmoMC-Nov2016_08/python/getdist/mcsamples.py", line 1047, in getAutoBandwidth1D
Neff = self._get1DNeff(par, param)
File "CosmoMC-Nov2016_08/python/getdist/mcsamples.py", line 1028, in get1DNeff
par.Neff_kde = self.getEffectiveSamplesGaussianKDE(param, scale=par.sigma_range)
File "CosmoMC-Nov2016_08/python/getdist/chains.py", line 407, in getEffectiveSamplesGaussianKDE
corr = np.zeros(maxoff + 1)
ValueError: Maximum allowed dimension exceeded

Can you please suggest a way out?
Mayukh
Back to top
View user's profile  
Mayukh Gangopadhyay



Joined: 04 Aug 2017
Posts: 6
Affiliation: Saha Institute of Nuclear Physics

PostPosted: August 25 2017  Reply with quote

Mayukh Gangopadhyay wrote:
Hi,
I have another problem regarding the same issue. I state it here:

After running with temperature = 2, when we analyse the data with cool = 2 in distparams.ini with "python python/GetDist.py distparams.ini", following is the error that we got:
Code:

Best fit sample -log(Like) = 1266.571000 
                                                                                                   
mean(-Ln(like)) = 1272.535056
                                                                                                                 
-Ln(mean like)  = 1268.978640 
                                                                                                               
Traceback (most recent call last):   
                                                                                                       
  File "python/GetDist.py", line 357, in <module>                                                                                             
    main(args)                                                                                                                               
  File "python/GetDist.py", line 194, in main                                                                                                 
    mc._setDensitiesandMarge1D(writeDataToFile=not no_plots and plot_data_dir, meanlikes=plot_meanlikes)                                     
  File "CosmoMC-Nov2016_08/python/getdist/mcsamples.py", line 2077, in _setDensitiesandMarge1D
    meanlikes=meanlikes)
  File "CosmoMC-Nov2016_08/python/getdist/mcsamples.py", line 1281, in get1DDensityGridData
    * (binmax - binmin) * abs(smooth_scale_1D) / fine_width
  File "CosmoMC-Nov2016_08/python/getdist/mcsamples.py", line 1047, in getAutoBandwidth1D
    N_eff = self._get1DNeff(par, param)
  File "CosmoMC-Nov2016_08/python/getdist/mcsamples.py", line 1028, in _get1DNeff
    par.N_eff_kde = self.getEffectiveSamplesGaussianKDE(param, scale=par.sigma_range)
  File "CosmoMC-Nov2016_08/python/getdist/chains.py", line 407, in getEffectiveSamplesGaussianKDE
    corr = np.zeros(maxoff + 1)
ValueError: Maximum allowed dimension exceeded


Can you please suggest a way out?
Mayukh
Back to top
View user's profile  
Antony Lewis



Joined: 23 Sep 2004
Posts: 1273
Affiliation: University of Sussex

PostPosted: August 25 2017  Reply with quote

Not sure what that is, but I'd try the latest getdist from github. If that still gives an error, please email me a link to your chains and analysis settings.
Back to top
View user's profile [ Hidden ] Visit poster's website
Mayukh Gangopadhyay



Joined: 04 Aug 2017
Posts: 6
Affiliation: Saha Institute of Nuclear Physics

PostPosted: August 25 2017  Reply with quote

Thanks will do that then.
Back to top
View user's profile  
Mayukh Gangopadhyay



Joined: 04 Aug 2017
Posts: 6
Affiliation: Saha Institute of Nuclear Physics

PostPosted: September 01 2017  Reply with quote

Hi,
I am still having trouble with the getdistGUI.
I can get the triangle plots only with getdistGUI, but cannot get the margestats and likestats files. The two main errors that it shows when trying to plot with getdistGUI or with python/getdist are following:
Code:

ValueError: Maximum allowed dimension exceeded
MCSamplesError: Parameter range is <= 0: chi2_prior

It would be great to get some advice
Back to top
View user's profile  
Antony Lewis



Joined: 23 Sep 2004
Posts: 1273
Affiliation: University of Sussex

PostPosted: September 01 2017  Reply with quote

For the "ValueError: Maximum allowed dimension exceeded " I don't know, I'd look at the call stack.

The range error is usually because you have something with a negative chi2. You can edit the .ranges file to set the range for that parameter not to start at 0.
Back to top
View user's profile [ Hidden ] Visit poster's website
Display posts from previous:   
Post new topic   Reply to topic    CosmoCoffee Forum Index -> Computers and software All times are GMT + 5 Hours
Page 1 of 1

 
Jump to:  
You cannot post new topics in this forum
You cannot reply to topics in this forum
You cannot edit your posts in this forum
You cannot delete your posts in this forum
You cannot vote in polls in this forum


Powered by phpBB © 2001, 2005 phpBB Group. Sponsored by WordWeb online dictionary and dictionary software.