MGT is a novel MaxSAT-based framework for solving the decoding phase of Group testing. This tool is based on our AAAI-20 paper.
The scripts are compatible with Python 3 or higher. The tool uses MaxHS as the underlying MaxSAT solver (checkout version c8df41ac63e27c2cb32905302e6ede3118b7a07b). MaxHS requires CPLEX to process the arithmetic computation of the MaxSAT problem. To setup the Python API of CPLEX, please follow the instruction from here.
mgt_benchmarks.py contains the subroutine main_varying_maxhs to generate and solve a group testing instance given the number of items n, probability of defective items p, type of tests (noiseless or noisy), and probability of noisy tests (in case of noisy test). This subroutine calls a set of supporting subroutines from group_testing_function.py, max_sat_interface.py, and exp_script.py. More specifically, group_testing_function.py contains useful scripts to generate the random item vector, the pool matrix, and to compute results after recovering the items. max_sat_interface.py acts as an interface to call a MaxSAT solver to solve the group testing instance. This script can be modified to use any off-the-shelf MaxSAT solver for solving the same problem. exp_script.py contains subroutines to show the experimental results as plots.
To compare MaxSAT-based framework with LP-relaxed (linear programming relaxation) approach, one would call the subroutine main_comparison_maxhs_lp from mgt_lp_comparison.py script. Moreover, general_lp_interface.py acts as an interface for calling a LP solver (the default choice is CPLEX) to solve the group testing instance.
Additionally, setup.py contains different choices including the directory of temporary files and generated plots.
To solve a noiseless group testing instance using MaxSAT, run the following commands.
import mgt_benchmark as mgt
import exp_script as plot
mgt.main_varying_maxhs(n=100, p=0.03, noiseless=True, verbose=True)
plot.parser(verbose=True)
The above commands solve a noiseless group testing instance for 100 items with 3% defective items.
To solve a noisy group testing instance with probability of noise as 5%, run the following command.
mgt.main_varying_maxhs(n=100, p=0.03, noiseless=False, noise_probability=0.05, verbose=True)
To compare the performance between the MaxSAT formulation and the LP formulation, run the following commands.
import mgt_lp_comparison as mgt_lp
import exp_script as plot
mgt_lp.main_comparison_maxhs_lp(n=100, p=0.03, noiseless=True, verbose=True)
plot.parser_comparison()
Similarly, to compare performance in the noisy group testing, run the following command.
mgt_lp.main_comparison_maxhs_lp(n=100, p=0.03, noiseless=False, noise_probability=0.05, verbose=True)
Run python test.py, that combines all above commands.
Lorenzo Ciampiconi ([email protected])
Bishwamittra Ghosh ([email protected])
@inproceedings{CGSM20,
author={Ciampiconi, Lorenzo and Ghosh, Bishwamittra and Scarlett, Jonathan and Meel, Kuldeep S.},
title={A {MaxSAT}-based Framework for Group Testing},
booktitle={Proceedings of AAAI},
year={2020},}