Dataset: Artificial collectives of specialists and generalists excel at different tasks

This dataset supports the following paper:

The accompanying code can be found in the following repository: https://github.com/meluso/mas-interpretive-networks-code

The dataset contains four studies. In each study, agents use a different optimization algorithm to search state spaces for high performing solutions. Files names are structured as aiteams01[optimizer]_[datasubset].tar. Optimizer names can be any of the following:

nm: Nelder-Mead optimizer, results of which are shown in the main paper.

da: Simulated annealing optimizer, supplement

lb: L-BFGS-B optimizer, supplement

rw: Random walk optimizer, supplement

Data subset names can be either raw for the raw data in parquet format, or analysis for pre-processed data in parquet format. The raw data for this project are large (multiple gigabytes). If you wish to perform our pre-processing again from scratch, we recommend downloading only the raw tar file you intend to analyze to save storage space. If you wish to perform analyses with compiled datasets or regenerate the figures in the paper, we recommend downloading the analysis tar file(s) corresponding to the figures you intended to generate.

To generate the figures described in the paper from pre-processed data, download and extract the analysis datasets from the analysis tar archives. Place the the extracted data into a directory called data in the root (highest level) directory of the repository. Then, from the root directory, run python plots.py. Similarly, you can run new simulations of your own by running simulation.py or reperform the analyses from our raw simulation data with analysis.py.