Code and data for "First Steps Towards a Runtime Analysis When Starting With a Good Solution"

This upload accompanies the experimental part of the paper "First Steps Towards a Runtime Analysis When Starting With a Good Solution", accepted for publication to ACM Transactions on Evolutionary Learning and Optimization.

The source code is contained in file "generic-onell.zip", which is a snapshot of the original repository available on GitHub. This is a project that covers more than one paper. Reproducibility notes are contained in README in the archive.

The relevant raw data produced for the paper is:

• onemax-sqrt.json: optimization runs of various algorithms, mentioned in the paper, on the OneMax problem of various sizes n, starting at a distance sqrt(n).
• onemax-log.json: same, starting at a distance ln(n+1).
• onemax-1to21-unlimited.json: same for problem size n=2^22, starting at distances equal to powers of 2 ranging from 1 to 2^21.
• onemax-1to14-limited.json: same for a smaller range of distances (up to 2^14), featuring heavy-tailed algorithms with small beta values and an explicit upper limit on the population size.

The Python script build.py constructs figures for the paper, as well as matrices of p-values using the Wilcoxon rank sum test. The calls to perform are:

• python build.py onemax-sqrt.tex onemax-sqrt.sig n onemax-sqrt.json
• python build.py onemax-log.tex onemax-log.sig n onemax-log.json
• python build.py onemax-etc.tex onemax-etc.sig d onemax-1to21-unlimited.json onemax-1to14-limited.json

The results are also in the upload. The figure sources are used in the paper:

• onemax-sqrt.tex: the source for Fig.5.
• onemax-log.tex: the source for Fig.6.
• onemax-etc.tex: the source for Fig.7.

The p-value tables compare algorithms for settings matching vertical slices of the above figures. The format is rather human-readable. the corresponding files are:

• onemax-sqrt.sig(Table 4 in the paper displays the last group of values).
• onemax-log.sig (Table 5 in the paper displays the last group of values).
• onemax-etc.sig

Also, there are the runs of the three-distribution version of the (1+(λ,λ)) algorithm on OneMax, and the corresponding runs of the (1+1) EA, in the following files:

• onemax-3d-experimentA.json - for Fig.8.
• onemax-3d-experimentB.json - for Fig.9.