Data for the paper, "Evolution of swarming behavior is shaped by how predators attack"

All of the data and the IPython Notebook we used to generate figures for our paper, "Evolution of swarming behavior is shaped by how predators attack" are contained in the included 'selfish-herd' directory. ------------------------------------------------------------------------------------------ If you have questions, contact us at: Randal S. Olson ([e-mail]) David B. Knoester ([e-mail]) ------------------------------------------------------------------------------------------ RUNNING IPYTHON NOTEBOOK To run our IPython Notebook, you must first download the required packages: * Python * IPython * IPython Notebook * glob * NumPy * SciPy * pandas We recommend downloading the Enthought Python distribution, which has all of the packages you will need. There is a free version for people with an active .edu email here: http://enthought.com/products/edudownload.php or a smaller, free version for those without an active .edu email here: http://enthought.com/products/epd_free.php Once you have installed the required packages, open your terminal and navigate to the eos-predator-confusion directory. Once there, enter the command: ipython notebook --pylab=inline This will open up the IPython Notebook terminal. Select the notebook labeled "Evolution of swarming behavior is shaped by how predators attack." This notebook is entirely interactive, so feel free to explore the data yourself. ------------------------------------------------------------------------------------------ DIRECTORY LABELS Each experiment has its own directory. 'kd' and the number following it represents the predator handling time, while 'np' and the number following it represents the number of predators in the experiment. The directory names are as follows: * swarm-sd30-selfish-herd-artificial-selection-combined-hdaa-and-oa-kd10-akd25-50kgens: Artificial selection experiments with 250 prey that experience both outside attacks and highest-density area attacks. These experiment directories have an 'akd' in the title, which represents the highest-density area attack predator's handling time. * swarm-sd30-selfish-herd-artificial-selection-outside-attack-kd4-50kgens: Artificial selection experiment with 250 prey and outside attacks. * swarm-sd30-selfish-herd-artificial-selection-outside-attack-probdeath-kd4-50kgens: Artificial selection experiment with 250 prey, outside attacks, and density-dependent predation. * swarm-sd30-selfish-herd-artificial-selection-rand-attack-kd4-50kgens: Artificial selection experiment with 250 prey and uniformly random attacks. * swarm-sd30-selfish-herd-artificial-selection-rand-attack-probdeath-kd4-50kgens: Artificial selection experiment with 250 prey, uniformly random attacks, and density-dependent predation. * swarm-sd30-selfish-herd-artificial-selection-randwalk-attack-kd4-50kgens: Artificial selection experiment with 250 prey and random walk attacks. * swarm-sd30-selfish-herd-artificial-selection-randwalk-attack-probdeath-kd4-50kgens: Artificial selection experiment with 250 prey, random walk attacks, and density-dependent predation. * swarm-sd30-selfish-herd-rand-starting-pos-density-dependent-survival-swarmsize100-np4-kd10: Predator-prey coevolution experiment with 100 prey and density-dependent predation. * swarm-sd30-selfish-herd-rand-starting-pos-swarmsize100-np4-kd10: Predator-prey coevolution experiment with 100 prey. ------------------------------------------------------------------------------------------ CSV FILE FORMAT Each csv file is one replicate for that experiment. The number following the word "run" in the csv file name is the random number generator seed for that replicate. e.g., run24LOD.csv has a random number generator seed of 24. In the LOD files, there will be a single entry for each ancestor in the final best swarm agent's LOD. LOD files will be in csv format with the column headers listed at the top. Column headers are in the following order: * generation: the generation the ancestor was born * prey_fitness: the fitness of the ancestor prey * predator_fitness: the fitness of the ancestor predator * num_alive_end: the number of surviving prey at the end of the fitness evaluation * avg_bb_size: the average bounding box size of the swarm during the simulation * var_bb_size: the variance in the bounding box size of the swarm during the simulation * avg_shortest_dist: the average distance from every prey agent to the prey agent closest to it * swarm_density_count: the average number of prey agents within safety distance units of each other * prey_neurons_connected_prey_retina: the number of conspecific sensor neurons that the prey Markov network is connected to * prey_neurons_connected_predator_retina: the number of predator sensor neurons that the prey Markov network is connected to * predator_neurons_connected_prey_retina: the number of prey sensor neurons that the predator Markov network is connected to * num_attacks: the number of attacks the predator made on prey during the simulation ------------------------------------------------------------------------------------------ OTHER FILES Some directories, such as 'swarm-sd30-selfish-herd-rand-starting-pos-swarmsize100-np4-kd10', also have .genome. Generally, we save Markov network files as .genome files. These files contain integer values which encode the Markov network. Other directories, such as 'swarm brains/swarm brains-oa-50kgens/', have .dot files, logic.txt, and minlogic.txt files. DOT files are the picture representations of Markov network structure and connectivity. We recommend using the Graphviz software to view these images. logic.txt files contain the truth table mapping every possible sensory input combination to the corresponding output from aor prey Markov network, as described in the paper. minlogic.txt files contain the truth table and representative logic of the prey Markov network, computed with the 'espresso' logic optimization software. ------------------------------------------------------------------------------------------

本研究论文《集群行为的演化受捕食者攻击方式影响》（Evolution of swarming behavior is shaped by how predators attack）所用的全部数据及用于生成论文图表的IPython 交互式笔记本（IPython Notebook），均收纳于附带的`selfish-herd`目录中。
------------------------------------------------------------------------------------------
若有相关疑问，请联系以下作者：兰德尔·S·奥尔森（Randal S. Olson，邮箱：[e-mail]）、戴维·B·克内斯特（David B. Knoester，邮箱：[e-mail]）
------------------------------------------------------------------------------------------
## IPython 交互式笔记本运行指南
若要运行本研究的IPython 交互式笔记本，请先安装所需依赖包：
* Python
* IPython
* IPython Notebook
* glob
* NumPy
* SciPy
* pandas
我们推荐使用Enthought Python发行版，该套件已包含上述全部所需依赖。针对拥有有效.edu邮箱的用户，可通过以下链接获取免费版本：http://enthought.com/products/edudownload.php；针对无.edu邮箱的用户，可通过以下链接获取精简免费版本：http://enthought.com/products/epd_free.php。
完成依赖包安装后，请打开终端并切换至`eos-predator-confusion`目录，随后执行如下命令：
`ipython notebook --pylab=inline`
此命令将启动IPython 交互式笔记本终端界面。请在界面中选择名为《Evolution of swarming behavior is shaped by how predators attack》的笔记本文件，该笔记本支持全交互操作，欢迎您自行探索相关数据。
------------------------------------------------------------------------------------------
## 目录命名规则
每项实验对应一个独立目录。以`kd`开头并紧随数字的字段，表示捕食者处理时长；以`np`开头并紧随数字的字段，表示实验中捕食者的数量。各目录的命名含义如下：
1. `swarm-sd30-selfish-herd-artificial-selection-combined-hdaa-and-oa-kd10-akd25-50kgens`：针对250个猎物个体的人工选择实验，猎物同时承受外部攻击与高密度区域攻击。该类实验目录名称中包含`akd`字段，用于表示高密度区域攻击捕食者的处理时长。
2. `swarm-sd30-selfish-herd-artificial-selection-outside-attack-kd4-50kgens`：针对250个猎物个体的人工选择实验，猎物仅承受外部攻击。
3. `swarm-sd30-selfish-herd-artificial-selection-outside-attack-probdeath-kd4-50kgens`：针对250个猎物个体的人工选择实验，猎物承受外部攻击且捕食存在密度依赖效应。
4. `swarm-sd30-selfish-herd-artificial-selection-rand-attack-kd4-50kgens`：针对250个猎物个体的人工选择实验，猎物承受均匀随机攻击。
5. `swarm-sd30-selfish-herd-artificial-selection-rand-attack-probdeath-kd4-50kgens`：针对250个猎物个体的人工选择实验，猎物承受均匀随机攻击且捕食存在密度依赖效应。
6. `swarm-sd30-selfish-herd-artificial-selection-randwalk-attack-kd4-50kgens`：针对250个猎物个体的人工选择实验，猎物承受随机游走式攻击。
7. `swarm-sd30-selfish-herd-artificial-selection-randwalk-attack-probdeath-kd4-50kgens`：针对250个猎物个体的人工选择实验，猎物承受随机游走式攻击且捕食存在密度依赖效应。
8. `swarm-sd30-selfish-herd-rand-starting-pos-density-dependent-survival-swarmsize100-np4-kd10`：针对100个猎物个体的捕食者-猎物共演化实验，捕食存在密度依赖效应。
9. `swarm-sd30-selfish-herd-rand-starting-pos-swarmsize100-np4-kd10`：针对100个猎物个体的捕食者-猎物共演化实验。
------------------------------------------------------------------------------------------
## CSV 文件格式说明
每个CSV文件对应一项实验的一次重复实验。文件名中`run`后的数字为该次重复实验的随机数生成器种子。例如，`run24LOD.csv`的随机数生成器种子为24。
在LOD文件中，最终最优集群智能体的每个祖先个体均对应一条记录。LOD文件采用CSV格式，首行为列标题，列标题的顺序及含义如下：
* `generation`：祖先个体诞生的世代数
* `prey_fitness`：祖先猎物的适应度
* `predator_fitness`：祖先捕食者的适应度
* `num_alive_end`：适应度评估结束时存活的猎物个体数
* `avg_bb_size`：模拟过程中集群的平均边界框尺寸
* `var_bb_size`：模拟过程中集群边界框尺寸的方差
* `avg_shortest_dist`：所有猎物智能体与其最近邻猎物智能体之间的平均距离
* `swarm_density_count`：彼此处于安全距离范围内的猎物智能体的平均数量
* `prey_neurons_connected_prey_retina`：猎物马尔可夫网络（Markov network）连接的同类感知神经元数量
* `prey_neurons_connected_predator_retina`：猎物马尔可夫网络（Markov network）连接的捕食者感知神经元数量
* `predator_neurons_connected_prey_retina`：捕食者马尔可夫网络（Markov network）连接的猎物感知神经元数量
* `num_attacks`：模拟过程中捕食者对猎物发起的攻击次数
------------------------------------------------------------------------------------------
## 其他文件说明
部分目录（如`swarm-sd30-selfish-herd-rand-starting-pos-swarmsize100-np4-kd10`）中包含`.genome`文件。通常，我们将马尔可夫网络（Markov network）文件保存为`.genome`格式，该文件以整数编码的形式存储马尔可夫网络结构。
另有部分目录（如`swarm brains/swarm brains-oa-50kgens/`）中包含`.dot`文件、`logic.txt`与`minlogic.txt`文件。其中，DOT文件为马尔可夫网络结构与连接关系的可视化图像，推荐使用Graphviz软件查看该类图像。`logic.txt`文件包含真值表，用于将所有可能的感知输入组合映射至猎物马尔可夫网络的对应输出，与论文中描述的内容一致。`minlogic.txt`文件包含经`espresso`逻辑优化软件处理后的真值表与简化后的猎物马尔可夫网络逻辑表达式。