jylls/ModelSelection_MonteCarlo_SubCompleteSpinModels

This Monte Carlo simulation finds the best sub-complete model with m independent variables in a system of n random variables (with n > m). For this purpose, it travels through the class of equivalence of these models by performing gauge transformations on the dataset and selects the model with the largest maximum log-likelihood.

README

Requirements:

MC_Algo1_32bits.cpp uses the C++11 version of C++.

Usage

Input datafile

The input datafile must be a list of n-bit binary numbers, each bit representing a spin variable (binary variable). Each line is a datapoint. The file must have no space between the columns (the bits of the same number).

Ex. 4 first datapoints of an input file with n=10 spin variables:

      0001000000
      1000100001
      0000000100
      0000100000

Each column is a spin; Each line is a datapoint.

At maximum n=32: At maximum you can set the number of variables to n=32. The reason is that datapoints are treated by the program as integers encoded on 32 bits, the lowest bit being the last column of your datafile. The code itself could easily be extended to larger values of n. However the present algorithm is efficient only for small values of n, and have difficulties to find the global maximum for n larger than 25. The search for the best sub-complete models in larger systems would require an evolved version of this algorithm.

Specification at the beginning of the MC_Algo1_32bits.cpp file

At the beginning of the MC_Algo1_32bits.cpp file, you must specify:

• const string data_file_name: the location and name of your datafile;
• const int n: the total number n of spin variables in your datafile, n<32;
• and const int m: the chosen number m of operators on which the sub-complete model is based, m<n.

The two values of n and m fully specify the class of sub-complete models the program will explore.

You can modify:

• const string directory: the output directory;
• const string MC_fileOUT: the name of the output file instead of outputfile_name, the details of the MC steps will be printed in this file.

You can also play with the parameters for the MC:

• const int N_MCsample: total number of MC steps (including shuffling);
• const double beta: the inverse of the temperature.

Compiling

For compiling, use the command:

 g++ -std=c++11 -O3 MC_Algo1_32bits.cpp

You may want to use the option -O3 of g++ to turn on some optimisations. If -std=c++11 doesn't work, please try -std=c++0x instead. This creates an executable a.out, that can be run using:

 ./a.out

You might want to rename the executable using the -c option of g++. These two command lines are reminded at the beginning of the MC_Algo1_32bits.cpp file.