Switching between Numerical Black-box Optimization Algorithms with Warm-starting Policies - Reproduction artifacts

<p># Reproducibility instructions</p> <p>This document details the steps to reproduce the results presented in the paper &#39;Chaining of Numerical Black-box Algorithms: Investigating the Impact of Warm-Starting and the Switching Point&#39;</p> <p>## Algorithm code<br> The code for the individual algorithms and the switching routines is included in the &#39;dynas.zip&#39; folder. &nbsp;</p> <p>## Data collection<br> To collect the performance data for the static algorithms, the file &#39;data_collection.py&#39; can be used. This file runs all 5 static algorithm on the function suite used in the paper, using the &#39;ioh&#39; package. To run it, please modify the &#39;ioh_dir&#39; variable to a place where the data should be stored.&nbsp;<br> The resulting data is made available under the &#39;static.zip&#39; file.</p> <p>## Processing code<br> The code which processes the performance data into the list of usecases is part of the &#39;Processing_R&#39; notebook.&nbsp;<br> This results in a set of rds-files with the data in IOHanalyzer&#39;s internal format, and in a csv file (&#39;Split_performances_relative.csv&#39;) with the selected usecases.&nbsp;</p> <p>This csv file can then be read in using the &#39;visualization&#39; notebook, which can generate Figure 1 from the paper. This notebook can then be used to count the specific usecases, which is the data shown in Table 1. The csv-file containing the selected usecases (&#39;usecases.csv&#39;) is also generated here.</p> <p>The &#39;usecases&#39; csv file can be used in the &#39;Data_collection_switch&#39; notebook to run the dynamic configurations. This results in the &#39;Dynamic.zip&#39; data, which can again be read using IOHanalyzer in the &#39;Processing_R&#39; notebook. This then generates several csv-files, for both the static and dynamic case, containing the per-run hitting times (&#39;RT_samples.zip&#39;). These are then used in the &#39;Visualization&#39; notebook to generate figure 3 and 4, as well as table 2.&nbsp;</p> <p>## Countour plots<br> The code for generating Figure 5 is included in the end of the &#39;Visualization&#39; notebook.</p> <p>## Performance comparison plot<br> Figure 6 has been generated using IOHanalyzer web-interface (iohanalyzer.liacs.nl), by uploading the files in &#39;cma_bfgs.zip&#39;.&nbsp;</p> <p>## Switch point impact plot<br> Figure 7 has been generated using the &#39;split_experiment&#39; notebook</p>