Efficient Bayesian phase estimation using mixed priors

This code is provided to allow reproducibility of the results in the paper: E. van den Berg, "Efficient Bayesian phase estimation using mixed priors" arXiv:2007.11629.

The cache directory contains all pre-computed results. These can be plotted using the relevant Python scripts:

• generate_figure_2.py
• generate_figure_3ab.py
• generate_figure_3c.py
• generate_figure_3de.py
• generate_figure_3f.py
• generate_figure_4ab.py
• generate_figure_4c.py
• generate_figure_4d.py
• generate_figure_4e.py
• generate_figure_4f.py
• generate_figure_5a.py
• generate_figure_5b.py
• generate_figure_5d.py
• generate_figure_5e.py
• generate_figure_6.py

The tables in the paper can be generated using:

• generate_tables_1_2.py

The output of the generate figure scripts is one or more figures in PDF format, stored in the fig directory. Automatic cropping of the figures using pdfcrop has been disabled because it is highly system dependent. However, it can be enabled by uncommenting the relevant section of the exportFigure function in the generic.py file.

The generate-figure scripts load precomputed results from the cache directory and generate the plots. To reproduce the results themselves, delete the relevant experiment files cache/experiment_hybrid_<index>.dat (likewise for the files cache/experiment_transition_<index>.dat) and regenerate the data as follows:

1. Run make run_experiment_hybrid or manually compile the files (run_experiment_hybrid.c, bpe.c, and bpe_plugins.c)

2. Create the cache directory if it does not already exist

3. On Mac and Linux run the experiment and redirect the output to file: ./run_experiment_hybrid <index> > ./cache/experiment_hybrid_<index>.dat where <index> is the experiment index needed by the figure generation script.

4. Once all data files have been generated, run the figure generation script.

For missing cache/experiment_transition_<index>.dat files, compile the executable using make run_experiment_transition, and run it with output redirected to cache/experiment_transition_<index>.dat. This part of the code, as well as the generation of the test problems could be improved. We decided to leave the code as-is to avoid introducing errors or inconsistencies with the data that was already generated.

Some verification of the code or equations is provided by the verify* scripts. The verify_clib_fourier.py script first calls make to compile verify_clib_fourier.c and then runs this code for comparison with the python implementation of the Fourier representation in bpe.py.

The compress.py script truncates data files for problems in the 200 and 300 ranges, since only data from the last several iterations is actually used (aside from two specific problem instances used for plotting). We added this script to reduce the file size for these problems and make it easier to check out the repository. When locally regenerating the data it can optionally be truncated by running the compress.py script.