help - python pyanimats.py -h basic usage - pyanimats.py <output_dir> <tasks.yml> [options]

1. Make virtualenv in Python 3

   • Mac

       `mkvirtualenv -p which python3 pyanimats`
     

   • Linux

       `mkvirtualenv -p /usr/bin/python3 pyanimats`
     

2. pip install -r requirements.txt

   (make sure gcc is installed because you need gfortran compiler for scipy: brew install gcc)

3. Rebuild each time any c++ changes

   python setup.py build_ext --inplace

1. Prepare a task as a yaml file, example:

   # tasks/1-3-1-3.yml
   # PyAnimat goals and tasks.
   # 1 means catch the block, -1 means avoid it.
   ---
   # 1-3-1-3 task.
   - [ 1, '1000000000000000']
   - [-1, '1110000000000000']
   - [ 1, '1000000000000000']
   - [-1, '1110000000000000']

2. Remove old build, and then rebuild it each time any c++ files change, (eg animats/constants.hpp)

   rm -r build && python setup.py build_ext --inplace

3. Run it with 30,000 generations, and sample 600 individuals from the evolution

   python pyanimats.py ./raw_results/0.0.21/initial_tests/seed-0 ./tasks/1-3-1-3.yml -n 30000 -i 600

In your selected output dir, will appear the following files:

• config.json

  The configuration that this game was run with, including for example: NUM_MOTORS, TASKS, NUM_NODES, MUTATION_PROB, etc.

• hof.pkl - "Hall of Fame"

  Individual's animat

• lineages.pkl

• logbook.pkl

• metadata.json

  Contains elapsed - seconds elapsed, and version - the version of the code

• pyanimats.py

  -i flag - total number of genomes to save in lineages.pkl make it use a db

• analyze.py

  make it use a db? except Larissa and Billy don't like dbs

• plot.py

• make_pkl_for_pyphi.py - uses analyze.py

• Main Inputs

  config.py - a bunch of defaults that prob get overridden configure.py - reads config.json's constants.py - params.yml - obsolete/deprecated (?) tasks/*.yml - animat/constants.hpp -

• Main Outputs

  • config.json -
  • metadata.json - version should be in dir structure, but it's here too just in case
  • hof.pkl - documented by DEAP, rarely used
  • lineages.pkl - double nested tuple, eg ((lineage of final animat - eg list of Animats, which are genomes sorta), (), ...)
    • note, each generation is determined probabilistically from the last generation, based on each Animat's fitness. So if in gen 1 you have 100 Animats, gen 2 will be determined by a mutated version of the previous generation's Animats, and the animats chosen for mutation are selected probabilistically based on their fitness.
  • logbook.pkl - documented by DEAP, obj from DEAP, a book w chapters, each chapter is some stat, each line in the "book" is a generation

• Game.cpp

• HMM.cpp

• individual.py - wrapper around the imported c++ animat object, plus some convenience stuff, this gets acted upon by the GA

• animat.pyx - cython file, wraps agent.cpp, gets turned into an importable file

• Agent.cpp - interacts with Game.cpp

• animat.cpp - autogenerated version of animat.pyx, this gets taken and turned into built object that can be imported by python

###pretty much for development and should be ignored###

• misc:

  silver searcher instead of grep 42, 213 = start codon

• --jumpstart

  • doesnt work if there's a config.json supplied
  • more start codons reaaaly slows things down... why?

• installing from github, maintaining branch/commit in pip freeze > requirements.txt

  • pip install -e git+https://github.com/wmayner/pyphi.git@a408bad57ad6df6d3d1c41787c69cda77d49adac#egg=pyphi

• End to End tests

• "analyze" module for loading genomes into python objects that can be manipulated

• Tests - what can be tested

• Architecture plan

• Hard coded things are tagged with # TODO: hard-coded: <description>