/Generalized_fish_game

Primary LanguagePython

Generalized_fish_game

Intended for use with the MOEAFramework, Borg and pareto.py. Licensed under the GNU Lesser General Public License.

Stage 0 - Set up

  • Make sure you have the MOEAFramework. Download the MOEAFramework-*-Demo.jar file and copy it to this directory.

  • Download the Borg source code and move files in this directory.

  • Download Pareto.py and put it in this directory

  • Within this directory create a directory called Generalized and inside it, create directories objs, sets, cnstrs, and runtime

Stage 1 - Optimization

  • Run the parallel optimization by running "qsub fish_game_opt_batch_script.pbs"

  • Get reference set and result file by running "sh find_refSets.sh". Notice the epsilon values used here, if you'd like a richer set of solutions, you should reduce the epsilon values for each objective.

  • Visualize solutions on parallel axis plot by running parallel_coordinate.py

Stage 2 - Re-evaluation of solutions in other SOWs

  • Generate Latin Hypercube Sample of deeply uncertain parameters by running "sh sample_parameters.sh". I strongly recommend to replicate the analysis with the same SOWs used in the paper and skip this step (i.e. use the parameter_samples.txt file already found in this directory). If one chooses to replicate with a new sample, making the visualizations will probably need additional work, to identify which SOWs from the sample to highlight and which sampled SOW lead to deterministic extinction.

  • Within the Generalized directory, create two directories: resim_objs and resim_cnstr

  • Re-simulate all identified solutions in all generated SOW by running "qsub resimulate_fish_game.sh". This will resimulate all identified solutions in all sampled SOWs.

  • To visualize how the entire set of solutions shifts under different SOWs, run scatter_uncertain_SOW.py. This file evaluates all the SOWs sampled and sorts them into stable and unstable, as well as calculate their distance from the base SOW and the stability inequality. In the figure in the paper, the SOWs visualized are: Orange - Nearby stable SOW, Green - Most distant stable SOW, Red - Unstable without deterministic extinction, Brown - Unstable with deterministic extinction (furthest away from inequality). Specifically, these are SOWs: 1540, 2832, 1253, and 2951 using python indexing. If you generate your own sample, these will obviously be different. This script takes a while to run.

  • To assess how the re-evaluated solutions compare to those that would have been obtained had we known the SOW we were in (i.e. the "regret"), we need to perform the optimization again, inputing the equivalent SOW parameters each time. This needs to be repeated three times, each time editing re_optimize_fish_game.py to call the equivalent SOW: 1540, 2832, and 1253. Reoptimization in 2951 (or any SOW with deterministic extinction is pointless as the system collapses anyway). We also need to create copies of the same directory architecture as in our original optimization, in the form of, e.g., "Reoptimized/SOW/objs" etc. After editing the python script and creating all necessary directories, run "qsub fish_game_reoptimize_batch_script.pbs" and repeat three times (for each SOW).

  • Get the reference set and result file for the optimization to each new SOW by running "sh find_refSets_reoptimization.sh". If you're looking into different SOWs, you'd need to change the directory and file names.

  • To visualize these against the re-evaluated solutions, run scatter_uncertain_SOW_regret.py.

  • To visualize the regrets on a parallel axis plot, run parallel_coordinate_multiple_SOW.py.

  • To visualize how species collapse occurs with parametric changes run parametric_space_collapse.py.

Stage 3 - Robustness analysis

  • For the robustness analysis, each solution needs to be assessed on whether or not it meets the performance criteria in each of the sampled SOWs. The percent of SOWs where each and all criteria are met is then calculated. SOWs where extinction is deterministic are left out of this calculation. File calcRobustness.py performs this analysis.

  • To visualize the robustness of each solution in a parallel axis plot, run robustness_parallel_coordinate.py. This script will also highlight two policies, the policy most robust in meeting the NPV performance criterion and the policy most robust in meeting all performance criteria.

  • To visualize the implications of policy preference on the system, policy_trajectories.py visualizes the application of the two highlighted policies on system dynamics