Your Universal Cellular Automata

https://mybinder.org/v2/gh/riveSunder/yuca/gecco_2022_pages?urlpath=/proxy/5006/gecco_bokeh_app

Check test line coverage -> here

Where are the log files and assets?

I've cleaned up this repository to make for a lighter git clone and install. Check https://github.com/rivesunder/yuca_docs for log files and assets.

Quick Start

Installation

virtualenv /my/env_path --python=python3.8
source /my/env_path/bin/activate

Install requirements:

# use a shallow depth to avoid a big download 
git clone -d 1 2022-07-28  https://github.com/riveSunder/yuca.git
# or
#git clone --shallow-since 2022-07-28  https://github.com/riveSunder/yuca.git

cd yuca
pip install -e .
pip install  jupyter notebook

Run testing:

python -m testing.test_all

Evolving CA rules

The tag argument (-t or --tag) currently has important implications for evolution runs. Including 'orbium' or 'geminium' in the tag string is used to set the neighborhood kernel used by the CA. Only the kernels from Lenia CA Hydrogeminium natans and Orbium are supported right now, and if neither is included in the tag the neighborhood kernel parameters will be sampled from a random uniform distribution. The neighborhood kernel is currently not evolved (only the update function parameters are), so it's not recommended to rely on randomly initialized kernels (yet).

python -m yuca.evolve -b 64 -c 256 -d cuda:0 -g 20 -k 13 -l 3 -m 128 -p 16 -s 42 -t my_geminium_tag

Evolving mobile patterns

To evolve patterns 'pattern' must be included in the tag (-t). It also wouldn't make sense to have a batch size (-b) or replicates (-l) greater than 1, because CPPNs and CA dynamics are both ostensibly deterministic (although floating point precision errors do sometimes play a role).

You can evolve patterns by loading a config file (-cc or --ca_config arg) or a progress log from a CA evolution run (-i or --input_filepath), the former is recommended.

python -m yuca.evolve -b 1 -c 256 -cc ca_configs/orbium.npy -d cuda:0 -g 20 -m 129 -p 128 -s 42 -t orbium_pattern_search

Command line args to the `yuca.evolve` entry point

This information can be accessed by entering yuca.evolve --help

"-b", "--batch_size"
    type=int 
    default=64
    help="number of grid instances (vectorization)"

"-c", "--ca_steps", 
    type=int
    default=1024 
    help="number of ca steps to search for"

"-ca", "--ca_fn"
    type=str
    default="CA"

"-cc", "--ca_config"
    type=str
    default=None
    help="filename (or filepath) designating a ca_config to load"

"-d", "--device"
    type=str
    default="cpu"
    help="device to use (cpu, cuda, or cuda:x)"

"-dt", "--dtype"
    type=str 
    default="float32"
    help="set default dtype in torch"

"-e", "--selection_mode"
    type=int
    default=0
    help="selection mode: 0: truncation, 1: rand. tourney, 2: proportional"

"-f", "--env_fn"
    type=str
    default="HaltingWrapper"

"-g", "--generations"
    type=int
    default=32
    help="number of generations to train"

"-i", "--input_filepath"
    type=str
    default=None
    help="npy log file training curves etc."

"-k", "--kernel_radius"
    type=int
    default=13
    help="kernel radius. kernel shape will be 2r+1 by 2r+1)"

"-l", "--replicates" 
    type=int
    default=1
    help="number of replicates to use in get_fitness"

"-m", "--dim"
    type=int
    default=128
    help="grid x,y dimension (square edge length)"

"-p", "--population_size"
    type=int
    default=32
    help="number of individuals in population"

"-r", "--prediction_mode"
    type=int
    default=0
    help="prediction mode: 0-vanishing, 1-static end, 2-both"

"-s", "--seed"
    type=int
    nargs="+"
    default=13
    help="seeds to initialized PRNGs, can enter multiple integers separated by spaces"

"-t", "--tag"
    type=str
    default="pattern_search"
    help="string tag for identifying experiments"

"-v", "--conv_mode"
    type=str
    default="circular"
    help="padding mode to use, 'circular', 'reflect', or 'zeros'"

riveSunder/yuca