EvoFlow - Evolutionary algorithms for humans
You have just found EvoFlow
EvoFlow is a modern hardware accelerated genetic algorithm framework that recast genetic algorithm programing as a dataflow computation on tensors. Conceptually is very similar to how Tensorflow & Keras are approaching deep-learning so if you have experience with any of those you will feel right at home.
Under the hood, EvoFlow leverage Tensorflow or Cupy to provide hardware accelerated genetic operations. If you don't have a GPU, you can run EvoFlow on Google Colab or it will just work fine on your CPU.
Getting started in 30 seconds
- Install EvoFlow:
pip install evoflow - Head to our hello world notebook that shows you how to use EvoFlow to solve the classic MaxOne problem.
Deep-learning versus Evoluationary algorithms
Generally you want to use Deep-learning when the problem is continious/smooth and evoluationary algorithms when the problem is discrete. For example voice generation is smooth and solving (non-linear) equations is discrete.
Concretely this means that the fitness functions you use to express what constraint to solve are very similar to the loss functions in deep-learning except they don't need to be differentiable and therefore can perform arbitrary computation.
However the cost of fitness function increased expressiveness and flexibility compared to neural network loss is that we don't have the gradients to help guide the model convergence and therefore coverging is more computationaly expensive which is why having a hardware accelerated framework is essential.
Genetic Algorithm terminology
- Gene: atomic unit. Equivalent to a neuron in deep-learning.
- Chromosome: ordered list of gene(s).
- Genotype: collection of chromosome(s). Used when the problem requires to maximizes multiples fitness function at once.
- Population of x: collection of chromosomes or genotypes. That is what makes a Tensor.
- Generation: One round of evolution. Equivalent to an epoch in deep-learning.
- Fitness function: Function that evaluate how good/fit a given chromosome is. this is equivalent to the loss function in deep learning except it doesn't need to be differentiable and aim to be maximized.
EvoFlow Terminology
-
evoluationary op: Operation performed on a population of chromosome to make them evolve. Common ops includes various type of Chromosomal crossovers and Chromosomal mutations. Equivalent to deep-learning layers (e.g a convolution layer).
-
evoluationary model: Directed graph of evolutionary ops that is used to evolve the population. This is equivalent to a model architecture in deep-learning settings.
Disclaimer
This is not an official Google product.
1 maintainer
Metadata
1 maintainer
Release files for evoflow 0.3.1589590018
For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.
Built distribution (wheel)
| File | Interpreter | ABI | Platform | Reset |
|---|---|---|---|---|
| evoflow-0.3.1589590018-py3.8.egg | Legacy Egg format | - | - | Details |
Release files / evoflow-0.3.1589590018-py3.8.egg
| Download URL | evoflow-0.3.1589590018-py3.8.egg |
|---|---|
| Size | 146.1 kB |
| Tags | Egg |
|
SHA-256 checksum How to use checksums |
9407c9828c54c5dbd75f68cbc9eb5b5a5e5d18d541f7987697484d3150a65d9a
|
|
BLAKE2b-256 checksum How to use checksums |
74f55b2a9635829a0913982fcac73a63db767d20519d06256c25f08844493554
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
twine/3.1.1 pkginfo/1.5.0.1 requests/2.23.0 setuptools/41.2.0 requests-toolbelt/0.9.1 tqdm/4.46.0 CPython/3.8.2
|
