GAMBIT plotting tools

A collection of Python tools for making publication-quality profile likelihood and posterior plots. The tools are developed in the context of GAMBIT and GAMBIT-light (see gambitbsm.org), but can be used as standalone tools.

Installation

Simply install using pip:

pip install git+https://github.com/GambitBSM/gambit_plotting_tools.git

LaTeX text rendering

By default this package will attempt to use Matplotlib's functionality for rendering text using LaTeX. A warning will be printed if the libraries required for LaTeX rendering are not found. In that case it will fall back to using Matplotlib's own mathtext functionality, and to get some decent-looking plots you probably should tweak font sizes, etc.

Examples

1D profile likelihood plots

• Make a single 1D profile likelihood plot from the data in the hdf5 file example_data/results_run1.hdf5

  python example_1D_profile_like_hdf5.py
  

  This should give a plot looking like this:

  

  This example also illustrates how to add the small GAMBIT logo to the bottom-right corner of a plot.

2D profile likelihood plots

• Make a single 2D profile likelihood plot from the data in the hdf5 file example_data/results_run1.hdf5

  python example_2D_profile_like_hdf5.py
  

  This should give a plot looking like this:

  

  This example also illustrates how to add a contour drawn from a set of stored coordinates, e.g. from some digitised plot.

• Make multiple 2D profile likelihood plots in one go, combining the data from the hdf5 files example_data/results_run1.hdf5 and example_data/results_run2.hdf5

  python example_2D_profile_like_hdf5_multiple.py
  

• Make a single 2D profile likelihood plot from the data in the ascii file example_data/samples.dat

  python example_2D_profile_like_ascii.py
  

• Make a 2D profile likelihood plot where the profile likelihood surface is coloured according to some other dataset

  python example_2D_profile_like_color_data_hdf5.py
  

  This example should produce a plot like this:

  

Conditional confidence intervals

• Make a series of confidence intervals for the y variable, conditioned on the x variable being within a given range

  python example_conditional_profile_intervals_shaded_hdf5.py
  python example_conditional_profile_intervals_colored_hdf5.py
  

  These examples give plots looking like this:

  

  By setting the argument x_condition to x_condition="upperbound" or x_condition="lowerbound" the confidence intervals for the y variable are conditioned on the x variable being below or above the given value on the x axis:

  

1D posterior plots

• Make a single 1D posterior plot from the data in the hdf5 file example_data/results_multinest.hdf5

  python example_1D_posterior_hdf5.py
  

  This should give a plot looking like this:

  

2D posterior plots

• Make a single 2D posterior plot from the data in the hdf5 file example_data/results_multinest.hdf5

  python example_2D_posterior_hdf5.py
  

  This should give a plot looking like this:

  

• Make 2D posterior credible regions coloured according to the posterior max or mean of a third dataset

  python example_2D_posterior_color_data_hdf5.py
  

  The example should produce a plot like this:

  

Conditional credible intervals

• Make a series of credible intervals for the y variable, conditioned on the x variable being within a given range, or above/below a certain value

  python example_conditional_credible_intervals_shaded_hdf5.py
  python example_conditional_credible_intervals_colored_hdf5.py
  

  In these examples the condition on x is an upper bound, which gives results like this:

  

  The markers show the conditional posterior mean.

2D scatter plots

• Make a 2D scatter plot using data from the hdf5 file example_data/results_run1.hdf5, with points ordered and coloured according to different datasets

  python example_2D_scatter_hdf5.py
  

  The example should give a plot like this:

  

Utility scripts

There are also some commands for other common tasks besides plotting

Print confidence level table

print_confidence_level_table

Print dataset names

• Read the hdf5 file example_data/results_run1.hdf5 and print all the dataset names, ignoring datasets whose name begins with "metadata" or ends with "_isvalid"

print_dataset_names example_data/results_run1.hdf5 --ignore-startswith "metadata" --ignore-endswith "_isvalid"

This should give an output looking like this:

data/#NormalDist_parameters @NormalDist::primary_parameters::mu
data/#NormalDist_parameters @NormalDist::primary_parameters::sigma
data/#normaldist_loglike @ExampleBit_A::lnL_gaussian
data/LogLike
data/MPIrank
data/ModifiedLogLike
data/Runtime(ms) interloop
data/Runtime(ms) intraloop
data/Runtime(ms) totalloop
data/Runtime(ns) for #NormalDist_parameters @NormalDist::primary_parameters
data/Runtime(ns) for #normaldist_loglike @ExampleBit_A::lnL_gaussian
data/Suspicious Point Code
data/pointID
data/scanID
data/unitCubeParameters[0]
data/unitCubeParameters[1]

Get parameters of the highest-likelihood points

• Read the data group in the hdf5 file example_data/results_run1.hdf5 and print the input parameters for the 3 highest-likelihood points

print_high_loglike_points example_data/results_run1.hdf5 data 3

This should give the following output:

File:  example_data/results_run1.hdf5
Group: data

The 3 highest log-likelihood point(s):

LogLike: -2.53130026064170e+01
Parameters:
  NormalDist:
    mu: 2.08331783688065e+01
    sigma: 2.15049587260387e+00


LogLike: -2.53130028409013e+01
Parameters:
  NormalDist:
    mu: 2.08336541456091e+01
    sigma: 2.15096678343953e+00


LogLike: -2.53130031660756e+01
Parameters:
  NormalDist:
    mu: 2.08344819535121e+01
    sigma: 2.14981940988637e+00