(c) 2020 Brendon J. Brewer.

$ git clone --recursive https://github.com/eggplantbren/DNest5

This will pull in a submodule that I'm using.

GNU General Public License version 3.

• Boost
• Python 3 with NumPy, matplotlib, apsw
• SQLite3
• yaml-cpp

To compile DNest5, run the build script. This generates a Makefile and then runs make to compile everything.

$ ./build.py

By default, this compiles the StraightLine example from DNest4. The two executable files created are main, which does the sampling, and postprocess, which creates posterior samples and computes the marginal likelihood.

First run main, then run showresults.py in Python. Showresults calls postprocess, so you don't need to run postprocess on its own. It is safe to run showresults.py while main is still running.

Edit ModelType.hpp and change the #include and using lines to point to a different model. Then run ./build.py again.

At this time, all example models in the Example directory are implemented using inheritance from the UniformModel class template. These follow the convention where the exploration is done in terms of a coordinate system that makes the prior Uniform(0, 1) for every coordinate. The user/model writer has to write the member functions us_to_params to convert from the us (Uniform(0, 1) parameters) to the actual parameters, and log_likelihood in order to evaluate the likelihood function.

The configuration file options.yaml plays the role of DNest4's OPTIONS and its command line arguments.

All outputs are written to the output directory. The files are:

• dnest5.db: An SQLite3 database of output, similar to DNest4's sample.txt, sample_info.txt, and levels.txt, but combined into one binary file.
• figure?.pdf: Figures exported by showresults.py.
• posterior.db: An SQLite3 database of posterior samples. Does not store the samples themselves, but rather refers back to the appropriate rows in dnest5.db.
• posterior.csv: CSV file of posterior samples.
• results.yaml: YAML file (plain text) with marginal likelihood values and related things.