2 releases

0.1.1 May 31, 2023
0.1.0 May 28, 2023

#390 in Images

MIT license

215KB
1.5K SLoC

lenia_ca

This crate has the core functionality for simulating the Lenia system of cellular automata. For more comprehensive documentation, please visit the docs.rs page for this crate.

At the time of publishing of this repository and crate, docs.rs does not allow embedding of images (not easily, anyways) in the documentation, and as a result, below are a few graphics viewable on Github showing the working principle of the two types of Lenia that the lenia_ca crate is able to simulate.

The working principle for StandardLenia is the following:

  • Perform a convolution operation (implemented as a FFT-based convolution) between the channel and kernel of the convolution_channel
  • Each point/pixel's value is then passed into a growth_function of the convolution_channel.
  • The resulting points/pixels are then multiplied by the integration step dt and added onto the original values in the channel.
  • The resulting points/pixels are then clamped to be in range 0..1. This result is the next time-step of the channel, and would be used as the next iteration's channel values.

use set_kernel() to change how the kernel looks.

use set_growth_function() to set a specific growth function for the convolution result.

use set_dt() to change the integration-step of the simulation.

Image representation of the algorithm on GitHub

The working principle for ExpandedLenia is the following:

  • For each convolution_channel, perform a convolution operation (implemented as a FFT-based convolution) between a source channel and the convolution_channel's kernel. Notice how each convolution_channel takes input from only one channel.
  • For each convolution_channel, pass the convolution results into the growth_function of the convolution_channel.
  • For each channel, perform an elementwise multiplication between the corresponding convolution_channel results and weights of the channel
  • For each channel, perform a weighted-sum on the results of the weight-convolution multiplicated results.
  • For each channel, multiply the weighted-sum by the integration step dt and add it to the original values in the channel.
  • For each channel, clamp the resulting values to be in range 0..1. This result is the next time-step of the corresponding channel, and would be used as the next iteration's channel values.

Image representation of the algorithm on GitHub

use set_channels() to set the number of channels in the simulation.

use set_convolution_channels() to set the number of kernels and the associated growth functions.

use set_convolution_channel_source() to set the channel which will be convoluted by a particular kernel.

use set_kernel() to change how a convolution_channel's kernel looks like.

use set_growth_function() to set a specific growth function for the convolution result.

use set_weights() to set a channel's weights for the corresponding convolution channel results.

use set_dt() to change the integration-step of the simulation.

Dependencies

~6.5MB
~121K SLoC