Exploring the fractal simulosphere in cellular automata

Getting started

The main dependencies you'll need are skimage, jax, mpi4py, matplotlib, and their dependencies. jupyter and coverage are also required for running notebooks and tracking unit test coverage, respectively.

First make sure you have libopenmpi-dev and python3 installed on your Linux machine. I am assuming you have a Linux machine: some finagling may be required for setup on windows subsystem for Linux or MacOS.

1. Install system dependencies

sudo apt-get update && sudo apt-get install -y libopenmpi-dev python-dev python3-venv 

2. Make a virtuale environment

I use virtualenv to manage different Python environments, but you may use a different virtual environment manager for python.

virtualenv fracatal --python=python3.8
source fracatal/bin/activate

3. Install Python dependencies:

pip install -r requirements.txt
pip install -e . 

Run tests:

python -m tests.test_all

Running experiments with CPU parallelization.

Currently the best way to run an experiment is with fractal/scripts/mpi_sweep.py via the command line. For example:

python -m fracatal.scripts.mpi_sweep -r 300 -a experiment_tag -g 256 -p 128 -n orbium_unicaudatus -w 33 -nkr 13.01 -nmu 0.15513 -ns 0.01705 -xs 0.01716 -ndt 0.3065 -xdt 0.3095 -t 10.0 -v 0

The cli args for mpi_sweep are shown below.

optional arguments:
  -h, --help            show this help message and exit
  -n PATTERN_NAME, --pattern_name PATTERN_NAME
                        the name of the pattern to evaluate, e.g. orbium_unicaudatus. see patterns folder
  -a TAG, --tag TAG     a string tag to help you find your experiments
  -p PARAMETER_STEPS, --parameter_steps PARAMETER_STEPS
                        number of parameter steps to sweep, result will be on a p by p grid
  -r MAX_RUNTIME, --max_runtime MAX_RUNTIME
                        (wall) time limit for experiment. experiments will always (try to) complete at least one run.
  -t MAX_T, --max_t MAX_T
                        maximum accumulated simulation time (in time units)
  -w WORKERS, --workers WORKERS
  -nmu MIN_MU, --min_mu MIN_MU
                        min value for mu (growth/target function peak). if no max_mu is provided, min_mu is used throughout
  -xmu MAX_MU, --max_mu MAX_MU
                        max value for mu (growth/target function peak)
  -ns MIN_SIGMA, --min_sigma MIN_SIGMA
                        min value for sigma (growth/target function bell width). if no max_sigma is provided, min_mu is used throughout
  -xs MAX_SIGMA, --max_sigma MAX_SIGMA
                        max value for sigma
  -ndt MIN_DT, --min_dt MIN_DT
                        min value for simulation time step dt. used throughout if max_dt is not provided.
  -xdt MAX_DT, --max_dt MAX_DT
                        max value for step size dt
  -nkr MIN_KR, --min_kr MIN_KR
                        min value for kernel radius kr. can take float values. min_kr used throughout if max_kr not provided
  -xkr MAX_KR, --max_kr MAX_KR
                        max value for kernel radius kr. can take float values.
  -k0 K0, --k0 K0       native kr value, used to scale glider patterns. currently over-ridden by 13 for Orbium and 31 for Hydrogeminium patterns
  -g GRID_DIM, --grid_dim GRID_DIM
                        grid dimensions for simulation. this must be larger than 2*max_kr+3
  -v VERBOSITY, --verbosity VERBOSITY
                        pass --verbosity 1 to print out debugging info during experiment.

Testing summary

TOTAL 1136 878 302 22 20% coverage.txt