Hi, @davidrmiller. Thank you for this awesome repository. I've created a fork that might be worth checking out. To get the glimpse on how it looks like, you can check screenshot.png in the root directory.

The main idea is to render simulation in real time rather than in the form of video files. With real time render simulation feels more alive, debugging is easier, and it opens whole lot more possibilities to observe and interact with the simulation. I've also tried to preserve existing functionality, so my PR has no innovative ideas, it's almost pure cosmetic that can be turned of.

Here is a brief list of changes:

• Added a new output using SFML which renders individuals, challenge criterias and barriers in real-time with the ability to drag simulation (LMB), zoom in/out (scroll), and select indiv (LMB)
• Added UI to watch and control simulation with TGUI library. Here is most notible controls:
  • pause/resume button, pause at start/end of generation buttons
  • component to slow or increase speed of simulation
  • components for changing some settings
  • button to restart the simulation
  • button to save selected indiv's net to svg
  • button to show indivs who would pass survival criteria if simulation would end now
• Added the ability to save (manual or auto) and load simulations with "cereal" serialization library

But first I've done some refactoring to prevent codebase from bloating because of the amount of additional files and to make it more suitable to the new flow.

• In simulator.cpp the entry point (simulator function) is used for top-level initialization and main application loop. Main simulation loop was extracted into new simulate function.
• New directory userio is grouping together everything related to input and output. Class UserIO is an entry point. It is called at specific points of simulation life cycle and passes it to SFMLUserIO and/or ImageWriter depending on which were enabled at the construction of UserIO. ImageWriter has not undergone any changes.
• SFMLUserIO is the main class for real life render. Display loop and event handling are done in this class. It also includes main window, view, all custom display components and child windows. All custom components are stored in sfmlComponents directory and created by SFMLUserIO with a heavy use of callback functions. Responsibility of custom components is to display UI, hey don't usually interact with simulation logic directly.
• survivalCriteria directory is where all challenges logic placed. It contains SurvivalCriteriaManager - a class that other parts of application are interacting with, and SurvivalCriteria successor classes that handles if certain indiv passes a criteria as well as displaying the criteria. The idea behind splitting criterias into dedicated classes was to make creating/disabling/enabling new challenges as easy as possible.
• ai directory contains everything related to genomes, indivs, signals, etc. No major changes here
• utils contains logic that not directly related to simulation, like: analysis, profiling tool, random number generator, and save/load system. save/load system uses cereal library which under the hood calls serialize function added for Params, Peeps and Indiv. Cereal library is located inside include library.
• All output files now go into Output directory. This includes:
  • Images for indiv's net
  • Logs
  • Profiling results
  • Save files of previous simulations

Application was tested on Linux Mint 21.3

New requirements are:

• sfml https://stackoverflow.com/questions/30696114/how-to-install-sfml-in-ubuntu
• tgui 1.1 https://tgui.eu/tutorials/latest-stable/

My solution is far from being perfect. UI needs more polishing, there are not enough space for all settings and buttons. Performance was not a first priority, so there is probably a lot that can be done on that front. Building inside a docker container also stopped working, I assume because it doesn't support SFML.