/nfoldexperiment

Evaluating and Tuning n-Fold Integer Programming

Primary LanguageJupyter Notebook

Evaluating and Tuning n-Fold Integer Programming

Supplementary material. For updates relating to the new submission, see the jea/ directory.

Code

Data

The format of the data is simple. Each file corresponds to a single run of the solver; each line corresponds to one outer loop (finding an augmenting step to apply according to the augmentation strategy), each number on a line corresponds to one run of the (AugILP) and is exactly the objective if this step would have been applied. The directory structure then corresponds to the various values of g1, augmentation strategy etc. (the directory names are self-explanatory).

Plots

We provide plots generated from data collected for scheduling instances with parameters m=15, lengths = {2,3,13,35}, weights = {6,13,2,1}, S=2000, L=10000, and slack ratio values 0.45, 0.50, 0.55, 0.60, 0.65. We have tested the augmentation strategies "best step", "2-apx best step", "5-apx best step" and "any step". The values of the parameter g1 were {2,3,4,5,6,7,8,9,10,12,14,16,18,20,23,26,29,33,40,45,50,60,75,90,110}.

Outer loop plots

Inner loop plots

Acknowledgement. Kateřina Altmannová is supported by the Student Faculty Grant "EXPERIMENTÁLNÍ ANALÝZA POKROČILÉHO ALGORITMU PRO N-FOLD IP". Dušan Knop is supported by projects NFR MULTIVAL and P202/12/G061 of GA ČR. Martin Koutecký is supported by Technion postdoctoral fellowship.