An (hopefully clean) implementation of Accelerated Neural Evolution through Cooperatively Coevolved Synapses in Python3
Right now I'm still implementing the base algorithm. The next phase will be profiling it to imporve its speed.
I'm done implementing the base structure, I've tested it on the rosenbrock function and it works ! I've done some profiling and the code now differs a bit from the paper's description but it's faster and does the same. I've made it a bit user friendly as well with a nice CLI.
The next phase will be to evaluate it on others problems using OpenAI's Gym. Also I should get it to work on multicore because right now something (not numpy and not the NeuralNetwork class) is limiting it to one core.
Also I should do something to make the evaluation method editable from the outside.
- cache: Preprocessed datasets that don’t need to be re-generated every time you perform an analysis.
- config: Configuration settings for the project
- data: Raw data files.
- preprocessing: Preprocessing data munging scripts, the outputs of which are put in cache.
- src: Statistical analysis and ML trainer scripts.
- diagnostics: Scripts to diagnose data sets for corruption or outliers.
- doc: Documentation written about the analysis.
- graphs: Graphs created from analysis.
- lib: Helper library functions but not the core statistical analysis.
- logs: Output of scripts and any automatic logging.
- profiling: Scripts to benchmark the timing of your code.
- reports: Output reports and content that might go into reports such as tables.
- tests: Unit tests and regression suite for your code.
- testing: Notebooks used for testing individual algorithm before definitive implementation.
- README.md: Notes that orient any newcomers to the project.
- TODO.md: list of future improvements and bug fixes you plan to make.