create pip-package (low prio for now)

[bjuergens]

long term goal: it should be easy to use this framework for others.

Currently this issue is not for specific implementation and instead for discussion about general ideas how to achieve that long term goal.

Most important question (from me):

• who is the target audience (besides us ourselves)
• how could users use or packages right now?
• do we need architectural changes?
• how should the package be called?

Also: I think we should rename this project. I think the name NeuroEvolution-CTRNN_new doesn't fit this project anymore

[bjuergens]

here some general steps, that I think may work towards this issue

• [ ] create setup.py, so it can be installed via pip
• [ ] polishing of interface to train.py
• [ ] reducing number of required parameters needed to start experiment
• [ ] make sure mp can deal with custom-gyms
  • [ ] idea: when gym is passed as a class (instead of an ID-string), then we could do the gym.register ourselves.
  • [ ] idea: The separation into main-process and sub-process (which currently happens in train.py with if __name__ == "__main__": should be moved to experiment.py.

a minimal example to start an experiment could look something like this:

import NeuroEvolution-CTRNN
import gym

class myEnv(gym.Env):
    pass 

experiment = NeuroEvolution-CTRNN.Experiment(population_size = 1000, generations = 200, env=myEnv)
experiment.start()
print("best score: " +str(experiment.log.max("fitness"))

[bjuergens]

also I think we should make sure we can handle plugins from the user. The user should be able to create custom brains, custom envs, custom optimizers and custom episode runners. I think custom brains and envs already works (albeit the definition of them is a little bit clunky at the moment).

[bjuergens]

@DanielZim reminder: you wanted to use a more object-oriented way to initialize experiment. Instead of passing configurations to the experiment, we could pass configurated objects for the brains, envs or optimizers. Also I noted, that for passing brains, we also had a related issue: #34