The implementations of the three PCAG solvers are in the same directory as this document. The directories AutoCCAG/, FastCA/ and TCA/ contains the implementation of AutoCCAG, FastCA and TCA, respectively.
sh build.sh
Note that the three PCAG solvers should be built on a 64-bit GNU/Linux operating system.
We provide simple_run.py, a Python3 script, as a unified interface for running these PCAG solvers. It allows the user to run AutoCCAG, FastCA or TCA with an initial PCA, which is either automatically generated by SamplingCA or explicitly provided by the user. For its detailed usage, users may execute it with the -h flag on.
Note that the initial PCA should be in the same format as the PCA constructed by SamplingCA.
Tip: The command for running SamplingCA is embedded in this script. Once there is unexpected error in running SamplingCA, users may refer to SamplingCA/README.md for solutions.
An example of using simple_run.py to run FastCA:
python3 simple_run.py -s F -i ../cnf_benchmarks/linux.cnf -o linux_PCA_by_FastCA.out --seed 1 --cutoff 20
The command above first calls SamplingCA to solve the instance ../cnf_benchmarks/linux.cnf with default hyper-parameter settings, since the initial PCA is not provided by the user. Then FastCA is called to optimize the initial PCA and the cutoff time is set to 20 seconds. The result is stored in linux_PCA_by_FastCA.out. Here both SamplingCA and FastCA use the random seed of 1.
The console output is expected to be similar with the following text:
c running SamplingCA ...
...
... <-- output of SamplingCA
...
c running FastCA ...
1.54254 110 0
3.49662 109 41
3.63432 108 46
4.08852 107 54
5.54425 106 110
8.32681 105 169
8.44476 104 173
13.2817 103 278
13.3068 102 279
c done.
The output between the lines c running FastCA ... and c done. is the output of FastCA. The second column represents the progress of PCA optimization, its size decreasing from 110 to 102. According to the console output, we expect the size of the PCA in linux_PCA_by_FastCA.out to be exactly 102.
The command is almost the same for running AutoCCAG and TCA, except that the argument -s (or its full name --solver) should take A for AutoCCAG and T for TCA. And the console output of AutoCCAG and TCA is also in similar format with FastCA.
We list the experimental results, which are related to the main competitors in this directory, as follows.
- Results_of_CAmpactor_and_its_SOTA_competitors.csv: Results of CAmpactor (with SamplingCA as its initialization algorithm) and its state-of-the-art competitors on all testing instances.
- Results_of_CAmpactor_on_its_generality.csv: Results of the generality of CAmpactor (with AutoCCAG, FastCA and TCA as its initialization algorithms) on all testing instances.
- Results_of_two_versions_of_AutoCCAG_FastCA_TCA.csv: Results of the versions of AutoCCAG, FastCA and TCA, which employ SamplingCA as their initialization algorithms, and the original versions of AutoCCAG, FastCA and TCA, which use their own, original initialization algorithms.