Navigation
Home Installing Running Download Version history Lists of samples SuperEGO
Contact the authors
Roberto Ruiz de Austri Roberto Trotta Farhan Feroz
Discussion forum
Webmaster - Please email me.

SuperEGO
SuperBayeS Enhanced Graphical Output

SuperEgo is a MATLAB Graphical User Interface which is used to analyze samples generated by SuperBayeS (either via MCMC or MultiNest). SuperEGO is intended mainly as an exploratory tool, and thanks to its interactivity is an ideal complement to getplots (which performs further statistical tests and generates MATLAB and SM scripts for the figures you want).

SuperEGO has been developed by Rachid Lemrani (CEA-Saclay) and it is based on the program CosmoloGUI developped by Sarah Bridle for CosmoMC.

This first release is currently a beta version which has not been fully tested yet. Please report problems with running or outputs to R. Lemrani (CEA-Saclay).

QuickStart guide

  • Download the SuperEgo scripts here .
  • Save the tarball in the SuperBayeS root directory, unzip and untar it (run "tar zxvf *.tgz")
  • Go into the SuperEGO directory generated, start MATLAB from there and in the MATLAB command prompt type superego

SuperEGO Screenshots and Instructions

Loading data

Once you have launched SuperEGO a window allows you to select the samples files you want to plot. Use the Add samples files to the list button to load several chains together (e.g., when you have several *_?.txt files generated by an MCMC run).

Locate and load the corresponding parameter names *.info file (produced by SuperBayeS in the same directory as the samples files).

SuperEGO then automatically loads the corresponding equal weights sample files (used for 2D and 3D scatter plots), namely *equal_weights.dat (for nested sampling samples) and *single.txt (for MCMC chains. This file needs to be generated by running getplots on the chains. If it is not found a warning message will appear).

Select the initial number of samples you want to discard (Burn In, useful for chains generated with MCMC. Set it to 0 if you are using nested sampling samples) and the Thin Factor (for a thin factor n, one every n samples is retained).


1D plotting

The controls should mostly be self-explanatory. The Hold button allows to superimpose on the same graph several 1D plots, e.g. marginal pdf, profile likelihood and mean likelihood distributions to compare them at a glance.

The 1D intervals (pdf) button turns on/off vertical lines showing the 68% and 95% (2-tails) credible (Bayesian) intervals from the marginal posterior.
The posterior mean button turns on/off a blue circle showing the mean from the marginal posterior.
The best fit button turns on/off a green triangle showing the location of the best-fit point.


2D plotting

2D plots of the marginal pdf, profile likelihood and mean likelihood distributions are available. 68% and 95% 2D joint contours can also be superimposed for the marginal pdf and for the profile likelihood. The colors drop-down menu allows to choose among a few color schemes. The absolute value of the quantities is usually not relevant.



3D plotting

3D plots can be produced using either the scatter option (to colour the scatter points by the value of a third variable) or a 3D surface showing the location of the 68% and 95% (marginal) hypersurface.

MATLAB 7.5.0 (R2007b) is known to generate out of memory errors on Mac OS X 10.5. Try using the pack command to free up some memory, or restart SuperEGO to reset the memory allocation.


Diagnostics plots

SuperEGO can be used to check convergence, mixing and burn in of the chains (mostly useful for MCMC generated samples).

Use variable name vs line numbers plots to check the chains movement in parameter space.

Use probability vs line numbers plots to assess the required burn in period (MCMC only).

There is also an implementation of the power spectrum test (currently beta version) to check samples independence (useful for both MCMC and nested sampling). For further details and interpretation, see astro-ph/0405462 .





Last updated on 1 June, 2010;  visits: