Inverse Ising

This repository contains Julia implementation of RISE, logRISE and RPLE algorithms for the inverse Ising problem.

Prerequisites

For running the code, you need to have the latest version of Julia installed on your computer, as well as JuMP, Ipopt and StatsBase packages.

Running

Specify desired parameters and file names in the arguments.csv file: reconstruction method (RISE, logRISE, RPLE), regularization coefficient c_lambda, symmetrization of reconstructed couplings (Y, N), name of the input sample file, name of the output parameters file.

Then run Inverse_Ising.jl in the command line:

julia Inverse_Ising.jl

The input csv sample file should be in the histogram form, where each line is in the format "number of time a configuration has been sampled, configuration". The output file of reconstructed parameters has the form of a csv matrix that inculdes couplings (as off-diagional entries) and magnetic fields (as diagonal entries).

For small systems (e.g. N<=25), samples can be exaustively generated with Gibbs_Sampler.jl:

julia Gibbs_Sampler.jl input_adjacency.csv num_samples output_samples.csv

See above for the formats of the adjacency matrix (containing parameters) and of the output file (containing samples in the histogram representation). A small synthetic example of input and output files is provided in the folder "synthetic_example".

D-Wave data set

The real data set generated on the D-Wave 2X quantum annealer "Ising" at Los Alamos National Laboratory and used in the paper for illustration is available in the folder "data_dwave".

Reference

If you find this code useful in your work, we kindly request that you cite the following paper:

A. Y. Lokhov, M. Vuffray, S. Misra, M. Chertkov (2016). Optimal structure and parameter learning of Ising models. arXiv preprint arXiv:1612.05024.

@article{lokhov2016optimal,
  title={Optimal structure and parameter learning of Ising models},
  author={Lokhov, Andrey Y and Vuffray, Marc and Misra, Sidhant and Chertkov, Michael},
  journal={arXiv preprint arXiv:1612.05024},
  year={2016}
}

License